Publikace

Artificial intelligence has seen several breakthroughs in recent years, with games often serving as milestones. A common feature of these games is that players have perfect information. Poker, the quintessential game of imperfect information, is a long-standing challenge problem in artificial intelligence. We introduce DeepStack, an algorithm for imperfect-information settings. It combines recursive reasoning to handle information asymmetry, decomposition to focus computation on the relevant decision, and a form of intuition that is automatically learned from self-play using deep learning. In a study involving 44,000 hands of poker, DeepStack defeated, with statistical significance, professional poker players in heads-up no-limit Texas hold'em. The approach is theoretically sound and is shown to produce strategies that are more difficult to exploit than prior approaches. © 2017, American Association for the Advancement of Science.

A complex system for control of swarms of micro aerial vehicles (MAV), in literature also called as unmanned aerial vehicles (UAV) or unmanned aerial systems (UAS), stabilized via an onboard visual relative localization is described in this paper. The main purpose of this work is to verify the possibility of self-stabilization of multi-MAV groups without an external global positioning system. This approach enables the deployment of MAV swarms outside laboratory conditions, and it may be considered an enabling technique for utilizing fleets of MAVs in real-world scenar- ios. The proposed visual-based stabilization approach has been designed for numerous different multi-UAV robotic applications (leader-follower UAV formation stabilization, UAVswarmstabilizationanddeploymentinsurveillancesce- narios, cooperative UAV sensory measurement) in this paper. Deployment of the system in real-world scenarios truthfully verifies its operational constraints, given by limited onboard sensing suites and processing capabilities. The performance of the presented approach (MAV control, motion planning, MAV stabilization, and trajectory planning) in multi-MAV applications has been validated by experimental results in indoor as well as in challenging outdoor environments (e.g., in windy conditions and in a former pit mine).

This work addresses classification of unknown binaries executed in sandbox by modeling their interaction with system resources (files, mutexes, registry keys and communication with servers over the network) and error messages provided by the operating system, using vocabulary-based method from the multiple instance learning paradigm. It introduces similarities suitable for individual resource types that combined with an approximative clustering method efficiently group the system resources and define features directly from data. This approach effectively removes randomization often employed by malware authors and projects samples into low-dimensional feature space suitable for common classifiers. An extensive comparison to the state of the art on a large corpus of binaries demonstrates that the proposed solution achieves superior results using only a fraction of training samples. Moreover, it makes use of a source of information different than most of the prior art, which increases the diversity of tools detecting the malware, hence making detection evasion more difficult.

Autonomous Micro Aerial Vehicles have the potential to assist in real life tasks involving grasping and transportation, but not before solving several difficult research challenges. In this work, we address the design, control, estimation, and planning problems for cooperative localization, grasping, and transportation of objects in challenging outdoor scenarios. We demonstrate an autonomous team of MAVs able to plan safe trajectories for manipulation of ferrous objects, while guaranteeing inter-robot collision avoidance and automatically creating a map of the objects in the environment. Our solution is predominantly distributed, allowing the team to pick and transport ferrous disks to a final destination without collisions. This result is achieved using a new magnetic gripper with a novel feedback approach, enabling the detection of successful grasping. The gripper design and all the components to build a platform are clearly provided as open-source hardware for reuse by the community. Finally, the proposed solution is validated through experimental results where difficulties include inconsistent wind, uneven terrain, and sandy conditions.

This paper describes a 3D vision system for robotic harvesting of broccoli using low-cost RGB-D sensors, which was developed and evaluated using sensory data collected under real-world field conditions in both the UK and Spain. The presented method addresses the tasks of detecting mature broccoli heads in the field and providing their 3D locations relative to the vehicle. The paper evaluates different 3D features, machine learning and temporal filtering methods for detection of broccoli heads. Our experiments show that a combination of Viewpoint Feature Histograms, Support Vector Machine classifier and a temporal filter to track the detected heads results in a system that detects broccoli heads with high precision. We also show that the temporal filtering can be used to generate a 3D map of the broccoli head positions in the field. Additionally we present methods for automatically estimating the size of the broccoli heads, to determine when a head is ready for harvest. All of the methods were evaluated using ground-truth data from both the UK and Spain, which we also make available to the research community for subsequent algorithm development and result comparison. Cross-validation of the system trained on the UK dataset on the Spanish dataset, and vice versa, indicated good generalisation capabilities of the system, confirming the strong potential of low-cost 3D imaging for commercial broccoli harvesting.

Autonomous systems will play an essential role in many applications across diverse domains including space, marine, air, field, road, and service robotics. They will assist us in our daily routines and perform dangerous, dirty, and dull tasks. However, enabling robotic systems to perform autonomously in complex, real-world scenarios over extended time periods (i.e., weeks, months, or years) poses many challenges. Some of these have been investigated by subdisciplines of Artificial Intelligence (AI) including navigation and mapping, perception, knowledge representation and reasoning, planning, interaction, and learning. The different subdisciplines have developed techniques that, when re-integrated within an autonomous system, can enable robots to operate effectively in complex, long-term scenarios. In this letter, we survey and discuss AI techniques as "enablers" for long-term robot autonomy, current progress in integrating these techniques within long-running robotic systems, and the future challenges and opportunities for AI in long-term autonomy.

In densely populated-cities, the use of private cars for personal transportation is unsustainable, due to high parking and road capacity requirements. The mobility-ondemand systems have been proposed as an alternative to a private car. Such systems consist of a fleet of vehicles that the user of the system can hail for one-way point-to-point trips. These systems employ large-scale vehicle sharing, i.e., one vehicle can be used by several people during one day and consequently, the fleet size and the parking space requirements can be reduced, but, at the cost of a non-negligible increase in vehicles miles driven in the system. The miles driven in the system can be reduced by ridesharing, where several people traveling in a similar direction are matched and travel in one vehicle. We quantify the potential of ridesharing in a hypothetical mobility-on-demand system designed to serve all trips that are currently realized by private car in the city of Prague. Our results show that by employing a ridesharing strategy that guarantees travel time prolongation of no more than 10 minutes, the average occupancy of a vehicle will increase to 2.7 passengers. Consequently, the number of vehicle miles traveled will decrease to 35 % of the amount in the MoD system without ridesharing and to 60% of the amount in the present state.

Cooperative multi-agent planning (MAP) is a relatively recent research field that combines technologies, algorithms, and techniques developed by the Artificial Intelligence Planning and Multi-Agent Systems communities. While planning has been generally treated as a single-agent task, MAP generalizes this concept by considering multiple intelligent agents that work cooperatively to develop a course of action that satisfies the goals of the group.

We present a new approach to long-term mobile robot mapping in dynamic indoor environments. Unlike traditional world models that are tailored to represent static scenes, our approach explicitly models environmental dynamics. We assume that some of the hidden processes that influence the dynamic environment states are periodic and model the uncertainty of the estimated state variables by their frequency spectra. The spectral model can represent arbitrary timescales of environment dynamics with low memory requirements. Transformation of the spectral model to the time domain allows for the prediction of the future environment states, which improves the robot’s long-term performance in dynamic environments. Experiments performed over time periods of months to years demonstrate that the approach can efficiently represent large numbers of observations and reliably predict future environment states. The experiments indicate that the model’s predictive capabilities improve mobile robot localisation and navigation in changing environments.