doc. Ing. Michal Jakob, Ph.D.

Electric vehicle (EV) travel planning is a complex task that involves planning the routes and the charging sessions for EVs while optimizing travel duration and cost. We show the applicability of the multi-objective EV travel planning algo- 5 rithm with practically usable solution times on country-sized road graphs with a large number of charging stations and a realistic EV model. The approach is based on multi-objective A* search enhanced by Contraction hierarchies, optimal dimensionality reduction, and sub-optimal ϵ-relaxation tech10 niques. We performed an extensive empirical evaluation on 182 000 problem instances showing the impact of various algorithm settings on real-world map of Bavaria and Germany with more than 12 000 charging stations. The results show the proposed approach is the first one capable of performing 15 such a genuine multi-objective optimization on realistically large country-scale problem instances that can achieve practically usable planning times in order of seconds with only a minor loss of solution quality. The achieved speed-up varies from ∼ 11× for optimal solution to more than 250× for sub20 optimal solution compared to vanilla multi-objective A*.

Current mobility patterns over-rely on transport modes that do not benefit sustainable and healthy lifestyles. To explore the potential for active mobility, we conducted a randomized experiment aimed at increasing regular commuter cycling in cities. In designing the experiment, we teamed up with developers of the “Cyclers” smartphone app to improve the effectiveness of the app by evaluating financial and non-financial motivational features. Participants in the experiment were recruited among new users of the app, and were randomly assigned to one of four different motivational treatments (smart gamification, two variants of a financial reward, and a combination of smart gamification and a financial reward) or a control group (no specific motivation). Our analysis suggests that people can be effectively motivated to engage in more frequent commuter cycling with incentives via a smartphone app. Offering small financial rewards seems to be more effective than smart gamification. A combination of both motivational treatments—smart gamification and financial rewards—may work the same or slightly better than financial rewards alone. We demonstrate that small financial rewards embedded in smartphone apps such as “Cyclers” can be effective in nudging people to commute by bike more often.

Activity-based modelling is a modern agent-based approach to travel demand modelling, in which the transport demand is derived from the agent’s needs to perform certain activities at specific places and times. The agent’s mobility is considered in a broader context, which allows the activity-based models to produce more realistic trip chains, compared to traditional trip-based models. The core of any activity-based model is an activity scheduler – a software component producing sequences of agent’s daily activities interconnected by trips, called activity schedules. Traditionally, activity schedulers used to rely heavily on hard-coded knowledge of transport behaviour experts. We introduce the concept of a Data-Driven Activity Scheduler (DDAS), which replaces numerous expert-designed components and their intricately engineered interactions with a collection of machine learning models. Its architecture is significantly simpler, making it easier to deploy and maintain. This shift towards data-driven, machine learning based approach is possible due to increased availability of mobility-related data. We demonstrate DDAS concept using our own proof-of-concept implementation, perform a rigorous analysis and compare the validity of the resulting model to one of the rule-based alternatives using the Validation Framework for Activity-Based Models (VALFRAM).

The basic premise of existing route planning algorithms is that complete information required for computing optimum routes is available at the time of the search. With the increasing complexity of transport systems, such an assumption is no longer valid as transport service providers are not willing to disclose full information about certain aspects of their services for business sensitivity reasons. Therefore, new approaches capable of computing optimal routes while minimizing the required amount of information about the services are required. In this paper, we investigate the incomplete information route planning problem in the context of planning routes with charging for electric vehicles. We have formalized the problem as a resource-constrained shortest path problem with time-dependent edge costs; the costs are only partially known and their values can be obtained by querying external data sources. We propose an optimum algorithm solving this problem utilizing two interchanging phases built on a multi-objective A* algorithm. We evaluate the properties of the algorithm on a comprehensive suite of test scenarios based on real-world data and derive insights into the properties of this emerging route planning problem.

Market mechanisms are now playing a key role in the allocation and pricing of on-demand transportation services. In practice, most such services use posted-price mechanisms, where both passengers and drivers are offered a journey price which they can accept or reject. However, providers such as Liftago and GrabTaxi have begun to adopt a mechanism whereby auctions are used to price drivers. These latter mechanisms are neither posted-price nor classical double auctions and can instead be considered a hybrid mechanism. In this paper, we describe and study the properties of a novel hybrid on-demand transport mechanism. As these mechanisms require knowledge of passenger demand, we analyze the data-profit tradeoff as well as how the passenger and driver preferences influence mechanism performance. We show that the revenue loss revenue for the provider scales with root n log n for n passenger requests under a multi-armed bandit learning algorithm with beta-distributed preferences. We also investigate the effect of subsidies on both profit and the number of successful journeys allocated by the mechanism, comparing these with a posted-price mechanism, showing improvements in profit with a comparable number of successful requests.

Over the last two decades, route planning algorithms have revolutionized the way we organize car travel. The advent of electric vehicles (EVs), however, bring new challenges for travel planning. Because of electric vehicle limited range and long charging times, it is beneficial to plan routes, charging, and activities jointly and in the context of the whole day—rather than for single, isolate journeys as done by standard route planning approaches. In this work, we therefore present a novel approach to solving such a whole day mobility problem. Our method works by first preprocessing an energy-constrained route planning problem and subsequently planning the temporally and spatially constrained activities. We propose both an optimal algorithm for the day mobility planning problem and a set of sub-optimal speedup heuristics. We evaluate the proposed algorithm on a set of benchmarks based on real-world data and show that it is significantly faster than the previous state-of-the-art approach. Moreover, the speedups provide dramatic memory and time improvements with a negligible loss in solution quality.

A major challenge for large-scale deployment of electrical vehicles (EVs) is charging. In general, the number of EVs that can be charged can be increased either by physically expanding charging capacity or by better exploiting existing charging capacity. In this paper, we focus on the latter, exploring how advanced EV travel planning systems can be used to better align where and when EV charging happens with where and when charging capacity is available. Our novel travel planner enables EV users to plan their trips and EV charging in a way that meets their needs yet reflects charging availability. The core innovation of our approach is that we take a broader, multidestination perspective to EV travel planning - this gives our planning system more flexibility and scope for deciding when and where charging should happen and, consequently, enables a better alignment between the need for charging implied by the EV travel and the availability of charging. We evaluate our approach on an agent-based simulation of several medium-scale scenarios based on real-world data. The results confirm the benefits of our multi-destination approach, especially in scenarios in which charging service providers support upfront booking of charging slots.

Efficient allocation of charging capacity to electric vehicle (EV) users is a key prerequisite for large-scale adaption of electric vehicles. Dynamic pricing represents a flexible framework for balancing the supply and demand for limited resources. In this paper, we show how dynamic pricing can be employed for allocation of EV charging capacity. Our approach uses Markov Decision Process (MDP) to implement demand-response pricing which can take into account both revenue maximization at the side of the charging station provider and the minimization of cost of charging on the side of the EV driver. We experimentally evaluate our method on a real-world data set. We compare our dynamic pricing method with the flat rate time-of-use pricing that is used today by most paid charging stations and show significant benefits of dynamically allocating charging station capacity through dynamic pricing.

With the increasing prevalence of electric vehicles (EVs), the provision of EV charging is becoming a standard commercial service. With this shift, EV charging service providers are looking for ways to make their business more profitable. Dynamic pricing is a proven technique to increase revenue in markets with time-variant, heterogeneous demand. In this paper, we propose a Markov Decision Process (MDP)-based approach to revenue-maximizing dynamic pricing for charging service providers. We implement the approach using an ensemble of policy iteration MDP solvers and evaluate it using a simulation based on real-world data. We show that our proposed method achieves significantly higher revenue than methods utilizing flat-based pricing. In addition to achieving higher revenue for charging service providers, the method also increases the efficiency of allocation measured in terms of the total utilization of the charging station.

On-demand transport has been disrupted by Uber and other providers, which are challenging the traditional approach adopted by taxi services. Instead of using fixed passenger pricing and driver payments, there is now the possibility of adaptation to changes in demand and supply. Properly designed, this new approach can lead to desirable tradeoffs between passenger prices, individual driver profits and provider revenue. However, pricing and allocations known as mechanisms are challenging problems falling in the intersection of economics and computer science. In this paper, we develop a general framework to classify mechanisms in ondemand transport. Moreover, we show that data is key to optimizing each mechanism and analyze a dataset provided by a real-world on-demand transport provider. This analysis provides valuable new insights into efficient pricing and allocation in on-demand transport.

We propose a novel and challenging variant of trip planning problems – Whole Day Mobility Planning with Electric Vehicles (WDMEV). WDMEV combines several concerns, which has been so far only considered separately, in order to realistically model the problem of planning mobility with electric vehicles (EVs). A key difference between trip planning for combustion engine cars and trip planning for EVs is the comparatively lower battery capacity and comparatively long charging times of EVs – which makes it important to carefully consider charging when planning travel. The key idea behind WDMEV is that the user can better optimize his/her mobility with EVs, if it considers the activities he/she needs to perform and the travel required to get to the locations of these activities for the whole day - rather than planning for single trips only. In this paper, we formalize the WDMEV problem and propose a solution based on a label-setting heuristic search algorithm, including several speed-ups. We evaluate the proposed algorithm on a realistic set of benchmark problems, confirming that the whole day approach reduces the time required to complete one’s day travel with EVs and that it also makes it cheaper, compared to the traditional single-trip approach.

Increasing the adoption of cycling is crucial for achieving more sustainable urban mobility. Navigating larger cities on a bike is, however, often challenging due to the cities' fragmented cycling infrastructure and/or complex terrain topology. Cyclists would thus benefit from intelligent route planning that would help them discover routes that best suit their transport needs and preferences. Because of the many factors cyclists consider in deciding their routes, employing a multicriteria route search is vital for properly accounting for cyclists' route-choice criteria. A direct application of optimal multicriteria route search algorithms is, however, not feasible due to their prohibitive computational complexity. In this paper, we formalize a multicriteria bicycle routing problem and propose several heuristics for speeding up the multicriteria route search. We evaluate our method on a real-world cycleway network and show that speedups of up to four orders of magnitude over the standard multicriteria label-setting algorithm are possible with a reasonable loss of solution quality. Our results make it possible to practically deploy bicycle route planners capable of producing diverse high-quality route suggestions respecting multiple real-world route-choice criteria.

Activity-based models, as a specific instance of agent-based models, deal with agents that structure their activity in terms of (daily) activity schedules. An activity schedule consists of a sequence of activity instances, each with its assigned start time, duration and location, together with transport modes used for travel between subsequent activity locations. A critical step in the development of simulation models is validation. Despite the growing importance of activity-based models in modelling transport and mobility, there has been so far no work focusing specifically on statistical validation of such models. In this paper, we propose a six-step Validation Framework for Activity-based Models (VALFRAM) that allows exploiting historical real-world data to assess the validity of activity-based models. The framework compares temporal and spatial properties and the structure of activity schedules against real-world travel diaries and origin-destination matrices. We confirm the usefulness of the framework on three real-world activity-based transport models.

On-demand transport services in the form of dial-a-ride and taxis are crucial parts of the transport infrastructure in all major cities. However, not all on-demand transport services are equal: not-for-profit dial-a-ride services with coordinated drivers significantly differ from profit-motivated taxi services with uncoordinated drivers. In fact, there are two key threads of work on efficient scheduling, routing, and pricing for passengers: dial-a-ride services; and taxi services. Unfortunately, there has been only limited development of algorithms for joint optimization of scheduling, routing, and pricing; largely due to the widespread assumption of fixed pricing. In this paper, we introduce another thread: profit motivated on-demand transport services with coordinated drivers. To maximize provider profits and the efficiency of the service, we propose a new market mechanism for this new thread of on-demand transport services, where passengers negotiate with the service provider. In contrast to previous work, our mechanism jointly optimizes scheduling, routing, and pricing. Ultimately, we demonstrate that our approach can lead to higher profits and reduced passenger prices, compared with standard fixed price approaches, while also improving efficiency.

This paper describes the experience in the Barcelona/Catalonia region concerning the set-up of a transport resource management system for personal mobility. This has been done under the umbrella of the EU FP7 MyWay project. MyWay develops beyond the traditional journey planners, with mobile user services that facilitate personalised seamless integration of public and private transport modes into a single trip and make travelling around the city effortless, swift and pleasurable. All possible transport modes available are displayed to the user in an integrated fashion, thus encouraging the use of cleaner modes of transport. The approach gives priority to the “egocentric vision” of the user, offering a solution closest to his or her personal needs and preferences, making the frequent use of this personalized solutions a main contribution to the sustainability of urban transport, as an alternative to ideal but not always practical and user acceptable solutions.

In this chapter, we describe an open-source simulation testbed for emerging autonomic mobility systems, in which transport vehicles and other resources are automatically managed to serve a dynamically changing transport demand. The testbed is designed for testing and evaluation of various planning, coordination and resource allocation mechanisms for the control and management of autonomic mobility systems. It supports all stages of the experimentation process, from the implementation of tested control mechanisms and the definition of experiment scenarios through simulation execution up to the analysis and interpretation of the results. The testbed aims to accelerate the development of control mechanisms for autonomic mobility systems and to facilitate their mutual comparison using well-defined benchmark scenarios. We also demonstrate how it can be used to select the most suitable control mechanism for a specific use case and to approximate operational costs and initial investments needed to deploy a specific autonomic mobility system.

Activity-based models are a specific type of agent-based models widely used in transport and urban planning to generate and study travel demand. They deal with agents that structure their behaviour in terms of daily activity schedules: sequences of activity instances (such as work, sleep or shopping) with assigned start times, durations and locations, and interconnected by trips with assigned transport modes and routes. Despite growing importance of activity-based models in transport modelling, there has been no work focusing specifically on statistical validation of such models so far. In this paper, we propose a six-step Validation Framework for Activity-based Models (VALFRAM) that exploits historical real-world data to quantify the model's validity in terms of a set of numeric metrics. The framework compares the temporal and spatial properties and the structure of modelled activity schedules against real-world origin-destination matrices and travel diaries. We demonstrate the usefulness of the framework on a set of six different activity-based transport models.

Market mechanisms play a key role in allocating and pricing commuters and drivers in new on-demand transport services such as Uber, and Liftago in Prague. These services successfully use different mechanisms, which suggests a need to understand the behavior of a range of mechanisms within the context of on-demand transport. In this paper, we propose a double auction mechanism and compare its performance to a mechanism inspired by Liftago’s approach. We show that our mechanism can improve efficiency and satisfy key properties such as weak budget balance and truthfulness.

Although in principle, providing a fully multimodal journey planner supporting the planning with all types of transport services and their combinations is now algorithmically possible, providing such fully multimodal journey planning in practice remains elusive. This situation is largely caused by the fact a fully multimodal journey planner requires a wealth of detailed data about all transport services and these are often difficult and/or costly to obtain. In this paper, we present a novel, elegant approach to overcome this problem. Our approach is based on journey planner integration – it obtains a fully multimodal journey planning capability by interconnecting, in a smart way, multiple incomplete journey planners, each with only limited modal and/or geographical coverage. The integration relies on a planning metagraph, a simplified representation of the underlying transport system which can be build with only a minimum amount of data. We have successfully tested our approach in real-world conditions of the greater metropolitan area of Barcelona, confirming the feasibility of our approach to achieve high-quality fully multimodal plans with a reasonable response time.

Widespread adoption of smartphones and ubiquitous internet connectivity gives rise to new markets for personalized and efficient on - demand mobility services. To rigorously analyze new control mechanisms for these services, we introduce an open - source agent - based simulation testbed that allows evaluating the performance of demand - responsive transport schemes. In particular, our testbed provides a framework to compare both centralized and decentralized, static and dynamic passenger allocation and vehicle routing mechanisms under various conditions; including varying vehicle fleets, road network topologies and passenger demands. The testbed supports all stages of the experimental process; from the implementation of control mechanisms and the definition of experiment scenarios, through to simulation execution, analysis, and interpretation of results. Ultimately, our testbed accelerates the development of control mechanisms for emerging on - demand mobility services a nd facilitates their comparison with well - defined benchmarks. We illustrate our approach on an example simulation study of standard taxi and taxi sharing services in the area of Sydney, Australia.

We describe a novel computationally efficient method for the fine-grained analysis of accessibility in multimodal (urban) transport systems. In contrast to existing work, we use a full-detail representation of the transport system, including exact timetables and complete maps of road, footpath and cycleway networks, which enables more accurate accessibility assessment. Compared to existing work, our approach also calculates a wider range of location- and time-dependent transport accessibility metrics, including service frequency and the number of transfers for public transport and physical effort and elevation gain for cycling. Because it employs efficient single-origin multiple-destination shortest-path graph search techniques, our method is also faster than existing approaches relying on point-to-point search algorithms. We demonstrate our method on the city of Prague, showing the interactive frontend for location accessibility analysis as well as the application of our algorithms for comparing the city-wide transport accessibility before and after large-scale timetable change.

Even though the agent-based simulation modelling has become a standard tool in transport research, current imple- mentations still treat travellers as passive data structures, updated synchronously at infrequent, predefined points in time, thus failing to cover within-the-day decision making and negotiation necessary for cooperative behaviour in a dynamic transport system. Leveraging the fully agent-based modelling approach, we have built large-scale activity-based models of multimodal mobility covering areas up to thousands of square kilometres and simulating populations of up to millions of inhabitants of several European cities. Citizens are represented by autonomous, self-interested agents which schedule and execute their activities (work, shopping, leisure, etc.) and trips in time and space. Indi- vidual decisions are influenced by agent’s demographic attributes and modelled using the data from mobility surveys. The model is statistically validated against origin-destination matrices and travel diary data sets.

Ridesharing, i.e., the problem of finding parts of routes that can be shared by several travellers with different points of departure and destinations, is a complex, multiagent decision-making problem. The problem has been widely studied but only for the case of ridesharing using freely moving vehicles not bound to fixed routes and/or schedules – ridesharing on timetabled public transport services has not been previously considered. In this paper, we address this problem and propose a solution employing strategic multiagent planning that guarantees that for any shared journey plan found, each individual is better off taking the shared ride rather than travelling alone, thus providing a clear incentive to participate in it. We evaluate the proposed solution on real-world scenarios in terms of the algorithm’s scalability and the ability to address the inherent trade-off between cost savings and the prolongation of journey duration. The results show that under a wide range of circumstances our algorithm finds attractive shared journey plans. In addition to serving as a basis for traveller-oriented ridesharing service, our system allows stakeholders to determine appropriate pricing policies to incentivise group travel and to predict the effects of potential service changes.

Increasing the adoption of cycling is crucial for achieving more sustainable urban mobility. Nav- igating larger cities on a bike is, however, often challenging due to cities’ fragmented cycling infrastructure and/or complex terrain topology. Cyclists would thus benefit from intelligent route planning that would help them discover routes that best suit their transport needs and preferences. Because of the many factors cyclists consider in deciding their routes, employing multi-criteria route search is vital for properly accounting for cyclists’ route-choice criteria. Dir- ect application of optimal multi-criteria route search algorithms is, however, not feasible due to their prohibitive computational complexity. In this paper, we therefore propose several heuristice for speeding up multi-criteria route search. We evaluate our method on a real-world cycleway net- work and show that speedups of up to four orders of magnitude over the standard multi-criteria label-setting algorithm are possible with a reasonable loss of solution quality. Our results make it possible to practically deploy bicycle route planners capable of producing high-quality route suggestions respecting multiple real-world route-choice criteria.

As new markets for transportion arise, on-demand transport services are set to grow as more passengers seek affordable personalized journeys. To reduce passenger prices and increase provider revenue, these journeys will often be shared with other passengers. As such, new negotiation mechanisms between passengers and the service provider are required to plan and price journeys. In this paper, we propose a novel profit-aware negotiation mechanism: a multiagent approach that accounts for both passenger and service provider preferences. Our negotiation mechanism prices each passenger's journey, in addition to providing vehicle routing and scheduling. We prove a stability property of our negotiation mechanism using a connection to hedonic games. This connection yields new insights into the link between vehicle routing and passenger pricing. We also show via simulations the dependence of the service provider profit and passenger prices on the number of passengers as well as passenger demographics. In particular, our key observation is that increasing the number of passengers has the effect of increasing passenger diversity, which in turn increases the service provider's profit.

We present a web-based tool for a fine-grained analysis of the quality of public transport coverage. Employing an efficient graph-based transport network representation and a fast, modified Dijkstra-based journey planning algorithm, the tool calculates four public transport accessibility indices: journey duration, service frequency, the number of transfers, and a combined, overall index. Together, the indices give an accurate picture of the user-perceived accessibility by public transport in the area and time of interest.

New markets for personalized and efficient transport are creating a need for on-demand mobility services. To rigorously analyse new control mechanisms for these services, we introduce an open-source agent-based simulation testbed that allows users to eva- luate the performance of multi-agent, on-demand transport schemes. In particular, our testbed provides a framework to compare both centralized and decentralized, static and dynamic passenger allocation and vehicle routing mechanisms under various con- ditions; including varying vehicle fleets, road network topologies and passenger demands. Our testbed supports all stages of the experimental process; from the implementation of control mechanisms and the definition of experiment scenarios, through to simu- lation execution, analysis, and interpretation of results. Ultimately, the testbed accelerates the development of control mechanisms for emerging on-demand mobility services and facilitates their comparison using well-defined benchmarks.

We present an open-source simulation testbed for the development, comparison and analysis of on-demand transport services. The testbed is designed to evaluate the performance of agent-based, on-demand transport vehicle allocation and routing mechanisms; accounting for different vehicle fleets, road network topologies and transport demand structures. A wide range of metrics, including passenger waiting times, vehicle occupancy, or distance travelled, can be used for measuring system performance. In order to encourage the comparison of different allocation and routing mechanisms under standard conditions, the testbed comes with a predefined set of ready-to-use benchmarking scenarios.

Bicycle route planning is a challenging problem because of the diverse set of factors considered by cyclists in choosing their cycling routes. We provide a solution to this problem based on a formal model expressive enough to represent transport network features and cyclists’ preferences grounded in the studies of real-world bicycle route choice behaviour. Our solution employs the A* algorithm together with vectors of cost and heuristic functions – able to optimise routes for travel time, comfort, quietness, and flatness.We have implemented, practically deployed and experimentally evaluated our solution in the challenging setting of the city of Prague. The experiments confirmed that the planner is able to return high-quality plans in less than 250 milliseconds per query.

To properly account for a broad range of routechoice factors in bicycle route planning, a multi-criteria optimization framework is needed. Unfortunately, in contrast to other categories of routing problems, optimal multi-criteria search has not yet been developed for bicycle routing. In this paper, we address this gap and provide a multi-criteria formulation of the bicycle routing problem and an optimum multi-label correcting algorithm for finding a full set of Pareto routes. To reduce the potentially very large number of Pareto solutions, we introduce a route selection algorithm, based on hierarchical clustering, for extracting a small representative subset of Pareto routes. We empirically evaluate our approach on a real-world cycleway network. We explore the size and structure of the set of Pareto routes and demonstrate the capability of our method to generate a practical set of bicycle routes in realistic conditions.

We present an advanced journey planner designed to help travellers to take full advantage of the increasingly rich, and consequently more complex offering of mobility services available in modern cities. In contrast to existing systems, our journey planner is capable of planning with the full spectrum of mobility services; combining individual and collective, fixed-schedule as well as on-demand modes of transport, while taking into account individual user preferences and the availability of transport services. Furthermore, the planner is able to personalize journey planning for each individual user by employing a recommendation engine that builds a contextual model of the user from the observation of user's past travel choices. The planner has been deployed in four large European cities and positively evaluated by hundreds of users in field trials.

Contemporary maritime piracy presents a significant threat to global shipping industry, with annual costs estimated at up to US$7bn. To counter the threat, policymakers, shipping operators and navy commanders need new data-driven decision-support tools that will allow them to plan and execute counter-piracy operations most effectively. So far, the provision of such tools has been limited. In cooperation with maritime domain stakeholders, we have therefore developed AgentC, a data-driven agent-based simulation model of maritime traffic that explicitly models pirate activity and piracy countermeasures. Modeling the behavior and interactions of thousands of individually simulated vessels, the model is capable of capturing the complex dynamics of the maritime transportation system threatened by maritime piracy and allows assessing the potential of a range of piracy countermeasures. We demonstrate the what-if analysis capabilities of the model on a real-world case study of designing a new transit corridor system in the Indian Ocean. The simulation results reveal that the positive past experience with the transit corridor in the narrow Gulf of Aden does not directly translate to the vast and open waters of the Indian Ocean and that additional factors have to be considered when designing corridor systems. The agent-based simulation development and calibration process used for building the presented model is general and can be used for developing simulation models of other maritime transportation phenomena.

We solve the fully multimodal journey planning problem, in which journey plans can employ any combination of scheduled public transport (e.g., bus, tram and underground), individual (e.g., walk, bike, shared bike and car), and on-demand (e.g., taxi) transport modes. Our solution is based on a generalised time-dependent graph that allows representing the fully multimodal earliest arrival problem as a standard graph search problem and consequently using general shortest path algorithms to solve it. In addition, to allow users to express their journey planning preferences and to speed up the search process, flexible journey plan templates can be used in our approach to restrict the transport modes and mode combinations permitted in generated journey plans. We have evaluated our solution on a real-world transport network of the city of Helsinki and achieved practically usable search runtimes in the range of hundreds of milliseconds.

The increasing pervasiveness of information and communication technology (ICT) in transport systems changes the requirements on techniques and tools for transport simulation modelling. Novel ICT-powered responsive mobility services, such as real-time on-demand transport, are emph{interaction-rich} in a sense that they rely on frequent, ad hoc interactions between various entities of the transport system. These interactions have to be properly captured in the model if it is to accurately represent the dynamics of the modelled transport system. Unfortunately, existing modelling tools are not well suited for modelling interaction-rich transport systems. We have therefore developed a novel modular simulation framework designed specifically for modelling transport systems in which ad hoc interactions and decision making play an important role. The framework provides an extensible library of modelling elements based on a unifying ontology of agent-based modelling abstractions, a high-performance discrete-event simulation engine and suite of tools supporting real-world deployment and utilization of implemented models. By fully leveraging the conceptual foundation of multiagent systems, our framework provides flexibility and extensibility that is difficult to achieve by existing approaches. We demonstrate the applicability of the framework on the models of five distinct interaction-rich transport systems.

Agent-based modelling has gained popularity as a useful technique for obtaining insights into the behaviour of complex adaptive systems in various fields, including economics, sociology and ecology. With its complex interactions between routes, schedules and engagement strategies, transit through piracy-affected waters is a problem very well suited for the application of this powerful modelling approach. Within the AgentC project, we have been developing a data-driven, agent-based model of global maritime traffic explicitly accounting for the effects of maritime piracy. The model employs finite state machines to represent the behaviour of merchant, pirate and naval vessels. It accurately replicates global shipping patterns and approximates real-world distribution of pirate attacks. By conducting and analysing results from thousands of simulation runs, the developed model and related tools allow gaining qualitative and quantitative insights into complex relationships governing piracy risks and costs. Further on, we utilize the simulation to conduct what-if analysis of possible piracy counter-measures and evaluate their effectiveness. Because of their strong application potential, the model and the tools are currently considered by the U.N. International Maritime Organization for potential use in assessing future operational counter-piracy measures, including new transit corridors and extended group transit schemes.

Multi-agent coordination and planning techniques play a fundamental role in designing control systems for autonomous multi-UAV-based information collection. The most important categories of information collection are exploration, surveillance, search and tracking. We introduce each category of information collection and then focus on a specific problem - multi-UAV-based surveillance in complex environments where sensor occlusions can occur due to high buildings and other obstacles in the target area. We formalize the problem and propose a solution as a decomposition of the problem in two subproblems: the problem of single-area surveillance and the problem of allocating UAVs to multiple areas. We present and compare several algorithms for the subproblems and prove their optimality. Finally, we empirically evaluate the performance of all algorithms on a realistic simulation of aerial surveillance, built using the AGENTFLYframework, and compare it to theoretical estimates.

AgentPolis is a fully agent-based platform for modeling multi-modal transportation systems. It comprises a high-performance discrete-event simulation core, a cohesive set of high-level abstractions for building extensible agent-based models and a library of predefined components frequently used in transportation and mobility models. Together with a suite of supporting tools, textsc{AgentPolis} enables rapid prototyping and execution of data-driven simulations of a wide range of mobility and transportation phenomena. We illustrate the capabilities of the platform on a model of fare inspection in public transportation networks.

Contemporary maritime piracy presents a significant threat to the global shipping industry, with annual costs estimated at up to US$12bn. To address the threat, commanders and policymakers need new data-driven decision-support tools that will allow them to plan and execute counter-piracy activities most effectively. So far, however, the provision of such tools has been very limited. To fill this gap, we have employed the multi-agent approach and developed a novel suite of computational tools and techniques for operational management of counter-piracy operations. A comprehensive agent-based simulation enables the stakeholders to assess the efficiency of a range of piracy counter-measures, including recommended transit corridors, escorted convoys, group transit schemes, route randomization and navy patrol deployments. Decision-theoretic and game-theoretic optimization techniques further assist in discovering counter-measure configurations that yield the best trade-off between transportation security and cost. We demonstrate our approach on two case studies based on the problems and solutions currently explored by the maritime security community. Our work is the first integrated application of agent-based techniques to high-seas maritime security and opens a wide range of directions for follow-up research and development.

A number of real-world security scenarios can be cast as a problem of transiting an area guarded by a mobile patroller, where the transiting agent aims to choose its route so as to minimize the probability of encountering the patrolling agent, and vice versa. We model this problem as a two-player zero-sum game on a graph, termed the transit game. In contrast to the existing models of area transit, where one of the players is stationary, we assume both players are mobile. We also explicitly model the limited endurance of the patroller and the notion of a base to which the patroller has to repeatedly return. Noting the prohibitive size of the strategy spaces of both players, we develop single- and double-oracle based algorithms including a novel acceleration scheme, to obtain optimum route selection strategies for both players. We evaluate the developed approach on a range of transit game instances inspired by real-world security problems in the urban and naval security domains.

An incremental process for developing human-agent-robot applications uses mixed-reality testbeds of varying fidelity and size.

SUPERHUB is a recently launched large-scale European effort aimed at enabling a new generation of eco-mobility services designed around and with the citizens. The SUPERHUB concept builds on the notion that citizens are not just mere users of mobility services, but represent an active component and a resource for policy-makers willing to improve sustainable mobility in smart cities. SUPERHUB makes use of innovative persuasive technology to raise awareness in citizens of the impact of their daily habits, fostering the adoption of virtuous, more environmentally-friendly mobility behaviours. From the technical standpoint, SUPERHUB is built around a big-data approach applied to mobility eco-systems, whereby data from a variety of sources (public transportation, road traffic information, GPS data, weather/pollution etc.) are collected and mined, with the help of advanced reasoning techniques and data analytics tools, in order to support citizens (through the provisioning of personalized and 'green' journey plans) as well as policy-makers (through the provisioning of detailed analysis of mobility flows and the assessment of the impact of mobility policies).

With the growing complexity of multi-agent applications and environments in which they are deployed, there is a need for development techniques that would allow for early testing and validation of application design and implementation. This is particularly true in cases where the developed multi-agent application is to be closely integrated with an existing, real-world system of multi-agent nature. Drawing upon our previous experiences with development of complex multi-agent applications, we propose simulation-aided design of multi-agent systems (SADMAS), a methodology tightly integrating simulations of the target system into the MAS application development process. In its heart lies the use of mixed-mode simulation, a simulation where parts of the deployed application operate in the target environment and parts remain simulated. We argue, that employing SADMAS process contributes to reduction of risks involved in development of complex MAS applications, as well as it helps to accelerate the process. Besides describing the capstones of the SADMAS approach and consequences of its application, we also illustrate it’s use on a case-study of a next-generation decentralised air traffic management system.

Recent rise in maritime piracy prompts the search for novel techniques for addressing the problem. We therefore developed AgentC, a prototype system that demonstrates how agent-based traffic management techniques can be used to improve the security of transit through piracy-affected areas. Combining agent-based modeling and simulation of maritime traffic and novel route planning and vessel scheduling techniques, the system shows the promising potential of agent-based methods for increasing maritime security.

Výzkumníci z Katedry kybernetiky Fakulty elektrotechnické na ČVUT studují použitelnost metod umělé inteligence a agentních technologií při zajišťování bezpečnosti v námořní doméně. Předmětem výzkumu sponzorovaného americkým Úřadem pro námořní výzkum (ONR) je vytvoření podrobného multiagentního modelu chování pirátů v prostředí námořní dopravy a návrh algoritmů pro optimálně znáhodněné plánování bezpečného transitu Adenským zálivem.

We study how a mobile defender should patrol an area to protect multiple valuable targets from being attacked by an attacker. In contrast to existing approaches we allow the targets to move through the area according to a priori known deterministic movement schedules. We represent the patrol area by a graph of arbitrary topology and do not put any restrictions on the movement schedules. We assume the attacker can observe the defender and has full knowledge of the strategy the defender employs. We construct a game-theoretic formulation and seek defender's optimal randomized strategy in a Stackelberg equilibrium of the game. We formulate the computation of the strategy as a mathematical program whose solution corresponds to an optimal time-dependent Markov policy for the defender.

Protection of privacy is a major concern for users of social web applications, including social networks. Although most online social networks now offer fine-grained controls of information sharing, these are rarely used, both because their use imposes additional burden on the user and because they are too complex for an average user to handle. To mitigate the problem, we propose an intelligent privacy manager that automates the assignment of sharing permissions, taking into account the content of the published information and user's high-level sharing policies. At the core of our contribution is a novel privacy policy language which explicitly accounts for social web concepts and which balances the expressive power with representation complexity. The manager employs named entity recognition algorithms to annotate sensitive parts of published information and an answer set programming system to evaluate user's privacy policies and determine the list of safe recipients. We implemented a prototype of the manager on the Face book platform. On a small test scenario, the manager reached the F-measure value of 0.831 in correctly recommending safe recipients.

Although most online social networks now offer fine-grained controls of information sharing, these are rarely used, both because their use imposes additional burden on the user and because there are too many control settings for an average user to handle. To mitigate this problem, we have developed an Intelligent Privacy Assistant for Facebook that partially automates the assignment of sharing permissions, taking into account the content of the information published and user's high-level sharing policies. The Assistant uses a novel social web privacy language, employs named entity recognition algorithms to annotate sensitive parts of published information and an answer set programming system to evaluate user's privacy policies and determine the list of safe recipients. On a test scenario, the Assistant reached 73.8% and 95.2% performance in correctly determining safe and unsafe recipients, respectively.

A number of real-world security scenarios can be cast as a problem of transiting an area patrolled by a mobile adversary, where the transiting agent aims to choose its route so as to minimize the probability of encountering the patrolling agent, and vice versa. We model this problem as a two player zero-sum game on a graph, termed the transit game. In contrast to the existing models of area transit, where one of the players is stationary, we assume both players are mobile. We also explicitly model the limited endurance of the patroller and the notion of a base to which the patroller has to repeatedly return. Noting the prohibitive size of the strategy spaces of both players, we employ iterative oracle-based algorithms including a newly proposed accelerated scheme, to obtain optimum route selection strategies for both players. We evaluate the developed approach on a range of transit game instances inspired by real-world security problems in the urban and naval security domains.

The recent surge in maritime piracy presents a serious threat to the international maritime transport system. Over the past few years, insurance rates have increased more than tenfold for vessels transiting known pirate waters, and the overall costs of piracy in the Pacific and Indian Oceans alone were estimated at up to US$16 billion in 2008 and continue to rise. To combat this problem, researchers have explored various measures for getting piracy back under control and for mitigating the risks it entails. Inspired by the recent successful applications of the multiagent approach to other traffic and transportation domains, we have developed a testbed for prototyping and evaluating agent-based techniques for understanding, detecting, anticipating, and eventually suppressing piracy and possibly other categories of maritime crime. Building on that testbed, we have investigated a range of specific coordination and planning methods for tackling the problem

Despite their use for modeling traffic in ports and regional waters, multi-agent simulations have not yet been applied to model mar- itime traffic on a global scale. We therefore propose a fully agent-based, data-driven model of global maritime traffic, focusing primarily on mod- eling transit through piracy-affected waters. The model employs finite state machines to represent the behavior of several classes of vessels and can accurately replicate global shipping patterns and approximate real- world distribution of pirate attacks. We apply the model to the problem of optimizing the Gulf of Aden group transit. The results demonstrate the usefulness of agent-based modeling for evaluating and improving op- erational counter-piracy policies and measures.

We present an agent-based system for modeling, analyzing and reasoning in the maritime domain with the emphasis on detecting, anticipating and preventing illegitimate activities, such as contemporary maritime piracy. At the core of the system is a data-driven agent-based simulation which combines a rich array of sources of real-world piracy-related data with simulated operation of thousands of vessels of different types in order to create a rich model of maritime activity. The simulation is integrated with a range of advanced adver- sarial reasoning methods for analyzing illegitimate activities and for planning active counter-measures. In combination with experiment support tools and a powerful presentation frontend based on Google Earth, the system provides a complete testbed for the development and evaluation of counter-piracy methods based on the multi-agent approach.

We explore how agent-based techniques can be employed to reduce the threat of contemporary maritime piracy to international transport. At the center of our approach is a data-driven agent-based simulation platform incorporating a range of real-world data sources in order to provide a solid computational model of maritime activity. The platform is integrated with extension modules providing advanced analysis, reasoning and planning capabilities. Two such modules are presented. The first module applies statistical machine learning techniques to extract models of vessel movement from trajectory data; the models are subsequently used for categorizing vessels and detecting suspicious activity. The second module employs game theory-based strategic reasoning to plan risk-minimizing routes for vessels transiting known pirate waters.

We present a novel approach to reducing adversarial search space by employing background knowledge represented in the form of higher-level goals that players tend to pursue in the game. The algorithm is derived from a simultaneous-move modification of the max n algorithm by limiting the search to those branches of the game tree that are consistent with pursuing player's goals. The algorithm has been tested on a real-world-based scenario modelled as a large-scale asymmetric game. The experimental results obtained indicate the ability of the goal-based heuristic to reduce the search space to a manageable level even in complex domains while maintaining the high quality of resulting strategies.

We present an agent-based coordination and planning method for autonomous aerial surveillance of multiple urban areas using a group of fixed-wing unmanned aerial vehicles (UAVs). The goal of the surveillance is to observe a set of ground points of interest within the target areas as often as possible. The method differs from the existing work by explicit consideration of sensor occlusions that can occur due to high buildings and/or other obstacles in the target area. The solution employs a decomposition of the problem in two sub- problems: the problem of single-area surveillance and the problem of allocating UAVs to multiple areas. The overall method is evaluated empirically on a realistic simulation of aerial surveillance built using the AgentFly framework.

We present an agent-based coordination and planning method for aerial surveillance of multiple urban areas using a group of fixed-wing unmanned aerial vehicles (UAVs). The method differs from the existing work by explicit consideration of sensor occlusions that can occur due to high buildings and other obstacles in the target area. The solution employs a decomposition of the problem in two subproblems: the problem of single-area surveillance and the problem of allocating UAVs to multiple areas. Three occlusion-aware methods for single-area surveillance are presented and compared. An algorithm for UAV allocation is presented and its optimality proved. The performance of all algorithms is evaluated empirically on a realistic simulation of aerial surveillance, built using the AgentFly framework, and is compared to theoretical estimates.

We present a method for learning characteristic motion patterns of mobile agents. The method works on two levels. On the first level, it uses the expectation-maximization algorithm to build a Gaussian mixture model of the spatial density of agents' movement. On the second level, agents' trajectories as expressed as sequences of the components of the mixture model; the sequences are subsequently used to train hidden Markov models. The trained hidden Markov models are then employed to determine agent type, predict further agent movement or detect anomalous agents. The method has been evaluated in the maritime domain using ship trajectory data generated by the AgentC maritime traffic simulation.

In today's world, we are increasingly surrounded by and reliant on complex sys- tems and infrastructures. Often, the se systems behave far from the optimum or even highly undesirable. Roads in our cities are congested, plane trips fre- quently delayed, computer networks rout inely overrun by worms and elektricity grids fail in split-second cascade react ions. Our systems have become massively interwoven and interdependent making both highly positive and negative chain reactions possible in critical systems. The systems that surround us, that provide us with communication, energy resources and support our safety and komfort are increasingly decentralized, interno nnected and autonom ous, with more and more decisions originating at the level of individual subsystems rather than be- ing imposed top-down. These systems are characterized by large numbers of geographically dispersed active entities with a complex network of mutual in- teractions and feedbacks, together giving rise to dynamic, non-linear emergent behavior which is very difficult to understand and even more difficult to control

We study the problem of a mobile agent trying to cross an area patrolled by a mobile adversary. The transiting agent aims to choose its route so as to minimize the probability of hostile encounter; the patroller agent, controlling one or more patrol units, aims at the opposite. We model the problem as a two-player zero-sum game (termed transit game) and search for an optimum route selection strategy as a mixed Nash equilibrium of the game. In contrast to existing game-theoretic models of this kind, we explicitly consider the limited endurance of patrols and the notion of bases to which the patrols need to repeatedly return. Noting the prohibitive size of the transit game, we employ two techniques for reducing the complexity of finding Nash equilibria - a compact network-flow-based representation of transit routes and iterative single- and doubleoracle algorithms for incremental game matrix construction

We introduce an adversarial planning algorithm based on game tree search, which is applicable in large-scale multiplayer domains. In order to tackle the scalability issues of game tree search, the algorithm utilizes procedural knowledge capturing how individual players tend to achieve their goals in the domain; the information is used to limit the search only to the part of the game tree that is consistent with pursuing players' goals. We impose no specific requirements on the format of the procedural knowledge; any programming language or agent specification paradigm can be employed. We evaluate the algorithm both theoretically and empirically, confirming that the proposed approach can lead to a substantial search reduction with only a minor negative impact on the quality of produced solutions.

We address the problem of multi-UAV surveillance in complex urban environments with occlusions. The problem consists of coordinating the flight of UAVs with on-board cameras so that the coverage and recency of the information about a designated area is maximized. In contrast to the existing work, sensing constraints due to occlusions and UAV flight constraints are modeled realistically and taken into account. We propose a novel occlusion-aware surveillance algorithm based on a decomposition of the surveillance problem into a variant of the three-dimensional art gallery problem and the multi-traveling salesmen problem for Dubins vehicles. The algorithm is thoroughly evaluated on the high-fidelity AGENTFLY UAV simulation testbed which accurately models all constraints and effects involved. The results confirm the importance of occlusion-aware flight path planning, in particular in the case of narrow street areas and low UAV flight altitudes.

We address the problem of multi-UAV surveillance in complex urban environments with occlusions. The problem consists of coordinating the flight of UAVs with on-board cameras so that the coverage and recency of the information about a designated area is maximized. In contrast to the existing work, sensing constraints due to occlusions and UAV flight constraints are modeled realistically and taken into account. We propose a novel occlusion-aware surveillance algorithm based on a decomposition of the surveillance problem into a variant of the three-dimensional art gallery problem and the multi-traveling salesmen problem for Dubins vehicles. The algorithm is thoroughly evaluated on the high-fidelity AGENTFLY UAV simulation testbed which accurately models all constraints and effects involved. The results confirm the importance of occlusion-aware flight path planning, in particular in the case of narrow street areas and low UAV flight altitudes.

Monitoring of fulfilment of obligations defined by electronic contracts in distributed domains is presented in this paper. A two-level model of contract-based systems and the types of observations needed for contract monitoring are introduced. The observations (inter-agent communication and agents' actions) are collected and processed by the contract observation and analysis pipeline. The presented approach has been utilized in a multi-agent system for electronic contracting in a modular certification testing domain.

Electronic contracting, based on explicit representation of different parties' commitments, is a promising way to specifying and regulating behaviour in distributed business applications. A key part of contract-based system is a process through which the actual behaviour of individual parties is checked for conformance with contracts set to govern such behaviour. Such checking requires that relevant information on the behaviour of the parties, both with respect to the application processes they execute and to managing their contractual relationships, is captured. The process of collecting all such information, termed contract observation, is the subject of this paper. First, we describe general properties and requirements of such an observation process; afterwards, we discuss specifics of realising contract observation in web services environments.

Electronic contracting, based on explicit representation of different parties's; commitments, is a promising way to specify and regulate behaviour in distributed business applications. In this paper, we show how the contract-based approach can be employed in the domain of European Computer Driving Licence (ECDL) testing in which test centres, testers and test room operators cooperate in order to deliver certification testing to ECDL candidates. We describe how the cross-organisational transactions involved in ECDL testing can be specified in terms of machine-processable business contracts and how these contracts can be automatically negotiated, managed and monitored using the technologies developed in the IST-CONTRACT project. The application of the contracting technology enables important business advantages, in particular increased flexibility and reliability in delivering the testing services to ECDL candidates.

We describe a distributed architecture for situated largescale agent-based simulations with predominately local interactions. The approach, implemented in AglobeX Simulation platform, is based on a spatially partitioned simulated virtual environment and allocating a dedicated processing core to the environment simulation within each partition. In combination with dynamic load-balancing, such partitioning enables virtually unlimited scalability of the simulation platform. The approach has been used to extend the AgentFly air-traffic testbed to support simulation of a complete civilian air-traffic touching National Air-Space of United States. Thorough evaluation of the system has been performed, confirming that it can scale up to a very high number of complex agents operating simultaneously (thousands of aircraft) and determining the impact of different configurations of the simulation architecture on its overall performance.

We present a novel approach to reducing adversarial search space by using background knowledge represented in the form of higher-level goals that players tend to pursue in the game. The algorithm is derived from a simultaneous-move modification of the maxn algorithm by only searching the branches of the game tree that are consistent with pursuing player's goals. The algorithm has been tested on a real-world-based scenario modelled as a large-scale asymmetric game. The experimental results obtained indicate the ability of the goal-based heuristic to reduce the search space to a manageable level even in complex domains while maintaining the high quality of resulting strategies.

Resources planning system for domain of modular certification testing is presented in this paper. In our domain, there are several independent resources owners and services providers, who together are able to provide their customers with service of certification testing. We are presenting concept of distributed system for supporting negotiation about the certification testing session's life-cycle support, i.e. from the session arrangement, through its execution, to results distribution. Each of the session participants is represented by mutually independent agents supporting all the certification session processes except the act of filling and evaluation of tests.

We study the problem of predictive data mining in a competitive multi-agent setting, in which each agent is assumed to have some partial knowledge required for correctly classifying a set of unlabelled examples. The agents are self-interested and therefore need to reason about the trade-offs between increasing their classification accuracy by collaborating with other agents and disclosing their private classification knowledge to other agents through such collaboration. We analyze the problem and propose a set of components which can enable cooperation in this otherwise competitive task. These components include measures for quantifying private knowledge disclosure, data-mining models suitable for multi-agent predictive data mining, and a set of strategies by which agents can improve their classification accuracy through collaboration. The overall framework and its individual components are validated on a synthetic experimental domain.

Of the ways in which agent behaviour can be regulated in a multi-agent system, electronic contracting - based on explicit representation of different parties' responsibilities, and the agreement of all parties to them - has significant potential for modern industrial applications. Based on this assumption, the CONTRACT project aims to develop and apply electronic contracting and contract-based monitoring and verification techniques in real world applications. This paper presents results from the initial phase of the project, which focused on requirements solicitation and analysis. Specifically, we survey four use cases from diverse industrial applications, examine how they can benefit from an agent-based electronic contracting infrastructure and outline the technical requirements that would be placed on such an infrastructure. We present the designed CONTRACT architecture and describe how it may fulfil these requirements.

Adaptability is a fundamental property of any intelligent system. In this paper, we present how adaptability in multi-agent systems can be implemented by means of collaborative logic-based learning. The proposed method is based on two building blocks: (1) a set of operations centred around inductive logic programming for generalizing agents' observations into sets of rules, and (2) a set of communication strategies for sharing acquired knowledge among agents in order to improve the collaborative learning process. Using these modular building blocks, several learning algorithms can be constructed with different trade-offs between the quality of learning, computation and communication requirements, and the disclosure of the agent's private information. The method has been implemented as a modular software component that can be integrated into the control loop of an intelligent agent.

Interactions and social relationships among agents are an important aspect of multi-agent systems. In this paper, we explore how such relationships and their relation to agent's objectives influence agent's decision-making. Building on the framework of stochastic games, we propose a classification scheme, based on a formally defined concept of interaction stance, for categorizing agent's behaviour as self-interested, altruistic, competitive, cooperative, or adversarial with respect to other agents in the system. We show how the scheme can be employed in defining behavioural norms, capturing social aspects of agent's behaviour and/or in representing social configurations of multiagent systems.

Adaptivity is a fundamental property of any intelligent system. In this paper, we present how adaptivity in multi-agent systems can be implemented by means of collaborative logic-based learning. The proposed method uses inductive logic programming to generalize agents' observations into sets of rules and inter-agent communication of acquired knowledge to improve team adaptation process. Based on Horn logic, the method is interoperable with semantic web languages and techniques. It has been implemented as a modular software component that can be readily integrated into the control loop of an intelligent agent or an autonomic component. We have evaluated the method on a realistic logistic scenario, in which teams of trading agents learn the properties of the environment in order to optimize their operation.

Service selection in an SOA is primarily based on matching functional requirements to those advertised by the services. The focus of this paper is on selecting, from those functionally capable services, the best service in terms of its non-functional attributes. We propose an adaptive multi-agent framework, Mercury, which involves collaborative modelling of the service landscape based on consumer experience of service quality. The framework introduces several mechanisms to enable the system to be adaptive, and thus effective in highly dynamic scenarios. In particular, different strategies of information exchange between selector agents are investigated with the aim of maximising the speed of adaptation while minimising unfavourable competition between consumers.

The paper describes a decentralized peer-to-peer multi-agent learning method based on inductive logic programming and knowledge trading. The method uses first-order logic for model representation. This enables flexible sharing of learned knowledge at different levels of abstraction as well as seamless integration of models created by other agents. A market-inspired mechanism involving knowledge trading is used for inter-agent coordination. This allows for decentralized coordination of learning activity without the need for a central control element. In addition, agents can participate in collaborative learning while pursuing their individual goals and maintaining full control over the disclosure of their private information. Several different types of agents differing in the level and form of knowledge exchange are considered. The mechanism is evaluated using a set of performance criteria on several scenarios in a realistic logistic domain extended with adversary behavior.

An important property of an autonomous agent is its ability to reason about other agents' private knowledge. Knowledge about the community and private knowledge of its members can be obtained by monitoring events appearing in the community. We propose three approaches how an agent can use the observed events to build such a model: one is based on theorem proving, the other on machine learning techniques, namely ILP and version space. The proposed framework allows to choose the type of model revision/update on eager-lazy strategy scale. These techniques were implemented and compared in two case studies concerning coalition formation.

An important type of knowledge representation is sets of symbolic IF.THEN rules. Symbolic rules are well understood by humans and amenable to symbolic manipulation and inference techniques. Rule extraction using neural networks presents an attractive approach to knowledge acquisition because it combines the straightforward manner in which neural networks can learn from training data with the above given advantages of rule sets. Rule extraction using neural networks proceeds by first training a neural network on the analyzed data, followed by transformation of the resulting network into a corresponding rule set representation. The article presents a rule induction technique that employ inversion of the network function in rule induction process. The technique is then generalized to a three-stage rule induction method and its advantages are discussed, including the use of decision region connectivity analysis for higher level-description of auxiliary middle model.

Steering behaviors present an important component of low-level control of autonomous agent in computer games, animation and robotics. The article extents existing steering behaviors with the \emph{directed seek} behavior, which steers the agent to arrive at a given waypoint in a given direction while considering minimum turn radius of the agent. In contrast to existing ad-hoc approaches, the article presents a geometrically well-founded solution which is robust, computationally efficient and compatible with other steering behaviors. Three-dimensional generalization of the algorithm as well as its several extensions are also presented. Directed seek behavior has been implemented and is tested as a part of the motion control system in a commercial action-strategy computer game.

Článek popisuje metodu získávání rozhodovacích pravidel z dat založenou na učení neuronové sítě jakožto mezikroku v extrakčním procesu.

The 2003 ECML/PKDD data mining challenge concerns a dataset describing the data collected during a longitudinal study of atherosclerosis prevention on around 1400 middle-aged men at a number of Czech hospitals. The data challenge entry from the Czech Technical University in Prague takes an approach which is heavy on data preparation through well-defined data transformations. This document describes the special requirements of this data mining tasks, the transformations designed to meet them and it points to some interesting observations found in the studied dataset.

The article describes a method for extracting decision rules from data by first training a neural network and then extracting rules from the learned network.