Ing. Jan Bayer

In this study, we address generalized autonomous mobile robot exploration of unknown environments where a robotic agent learns a traversability model and builds a spatial model of the environment. The agent can benefit from the model learned online in distinguishing what terrains are easy to traverse and which should be avoided. The proposed solution enables the learning of multiple traversability models, each associated with a particular locomotion gait, a walking pattern of a multi-legged walking robot. We propose to address the simultaneous learning of the environment and traversability models by a decoupled approach. Thus, navigation waypoints are generated using the current spatial and traversability models to gain the information necessary to improve the particular model during the robot's motion in the environment. From the set of possible waypoints, the decision on where to navigate next is made based on the solution of the generalized traveling salesman problem that allows taking into account a planning horizon longer than a single myopic decision. The proposed approach has been verified in simulated scenarios and experimental deployments with a real hexapod walking robot with two locomotion gaits, suitable for different terrains. Based on the achieved results, the proposed method exploits the online learned traversability models and further supports the selection of the most appropriate locomotion gait for the particular terrain types.

For autonomous robots operating in an unknown environment, it is important to assess the traversability of the surrounding terrain to improve path planning and decision-making on where to navigate next in a cost-efficient way. Specifically, in mobile robot exploration, terrains and their traversability are unknown prior to the deployment. The robot needs to use its limited resources to learn its terrain traversability model on the go; however, reusing a provided model is still a desirable option. In a team of heterogeneous robots, the models assessing traversability cannot be reused directly since robots might possess different morphology or sensory equipment and thus experience the terrain differently. In this paper, we propose a transfer learning approach for convolutional neural networks assessing the traversability between heterogeneous robots, where the transferred network is retrained using data available for the target robot to accommodate itself to the robot’s traversability. The proposed method is verified in real-world experiments, where the proposed approach provides faster learning convergence and better traversal cost predictions than the baseline.

The terrain traversal abilities of multi-legged walking robots are affected by gaits, the walking patterns that enable adaptation to various operational environments. Fast and lowset gaits are suited to flat ground, while cautious and highset gaits enable traversing rough areas. A suitable gait can be selected using prior experience with a particular terrain type. However, experience alone is insufficient in practical setups, where the robot experiences each terrain with only one or just a few gaits and thus would infer novel gait-terrain interactions from insufficient data. Therefore, we use knowledge transfer to address unsampled gait-terrain interactions and infer the traversal cost for every gait. The proposed solution combines gaitterrain cost models using inferred gait-to-gait models projecting the robot experiences between different gaits. We implement the cost models as Gaussian Mixture regressors providing certainty to identify unknown terrains where knowledge transfer is desirable. The presented method has been verified in synthetic showcase scenarios and deployment with a real walking robot. The proposed knowledge transfer demonstrates improved cost prediction and selection of the appropriate gait for specific terrains.

Predicting mobile robots' traversability over terrains is crucial to select safe and efficient paths through rough and unstructured environments. In multi-robot missions, knowledge transfer techniques can enable learning terrain traversability assessment the robots did not experience individually. The knowledge can be incrementally aggregated for homogeneous robots since they can treat foreign knowledge as their own. However, robots with different perceptions might experience the same terrain differently, so it is impossible to aggregate the shared knowledge directly. In this paper, we show how to learn a model that transfers the experience between heterogeneous robots, enabling each robot to use the whole sum of the experience of the multi-robot team. The proposed approach uses correlation to combine individual neural networks that assess the traversability of individual robots. The presented method has been verified in a real-world deployment of multi-legged walking robots with different cost assessment policies.

This paper concerns a mapping framework for multi-robot exploration of underground environments with only very limited communication available. We focus on multi-robot map building and coordination to explore large areas with real-time planning to long distances. The considered communication can broadcast only 100 B/s, and therefore, we propose coordination planning using two terrain models. The first model is a dense 3D map built by each robot individually to identify explorable places and generate detailed plans to avoid un-traversable areas. The second model is a global topological map built in a decentralized manner by exchanging tiny 12 B packets between the robots. The feasibility of the proposed approach has been verified in the real-world autonomous exploration mission and various multi-robot scenarios inspired by a virtual cave circuit of the DARPA Subterranean Challenge while adapting two different decentralized coordination strategies.

In this paper, we report on the experimental evaluation of the embedded visual localization system, the Intel RealSense T265, deployed on a multi-rotor unmanned aerial vehicle. The performed evaluation is targeted to examine the limits of the localization system and discover its weak points. The system has been deployed in outdoor rural scenarios at altitudes up to 20 m. The Absolute trajectory error measures the accuracy of the localization with the reference provided by the differential GPS with centimeter precision. Besides, the localization performance is compared to the state-of-the-art feature-based visual localization ORB-SLAM2 utilizing the Intel RealSense D435 depth camera. In both types of experimental scenarios, with the teleoperated and autonomous vehicle, the identified weak point of the system is a translation drift. However, taking into account all experimental trials, both examined localization systems provide competitive results.

The Subterranean Challenge (SubT) is a contest organised by the Defense Advanced Research Projects Agency (DARPA). The contest reflects the requirement of increasing safety and efficiency of underground search-and-rescue missions. In the SubT challenge, teams of mobile robots have to detect, localise and report positions of specific objects in an underground environment. This paper provides a description of the multi-robot heterogeneous exploration system of our CTU-CRAS team, which scored third place in the Tunnel Circuit round, surpassing the performance of all other non-DARPA-funded competitors. In addition to the description of the platforms, algorithms and strategies used, we also discuss the lessons-learned by participating at such contest.

In this paper, we address the problem of human safety during the search of underground or unknown indoor environments in scenarios where it is required to share the information about the position of the personnel such as first responders or speleologists. We report on the developed localization device to estimate and transmit the user position and state through a custom communication system. The information provided by the device serves to monitor the user state, enable communication between the user and the base station located outside the environment being searched. The personnel's position in the environment is thus provided to the mission supervisor in the case of an emergency. The device is based on low-cost off-the-shelf cameras for vision-based localization and low-bandwidth communication modules suitable for long-range communication in underground environments. Furthermore, the communication range is extendable by affordable transceivers that allow the dynamical building of independent communication networks in the operational environment. The performance of the communication system has been examined in an experimental scenario where five transceivers sufficiently cover 280 m of cave tunnels. The system can localize a human independently on any infrastructure supported by a report on the experimental deployment in an urban scenario with both indoor and outdoor areas.

In this paper, we address a problem of the exploration of large-scale subterranean environments using autonomous ground mobile robots. In particular, we focus on an efficient data representation of the large-scale elevation map, where it is desirable to capture the shape of the terrain to avoid areas not traversable by a robot. Subterranean environments such as mine tunnel systems can be in units of kilometers large, but only a relatively small portion of the environment represents observable parts. Therefore, uniform grid-based elevation maps with resolution in units of centimeters are not memory efficient, and more suitable are hierarchical tree-based structures. However, hierarchical structures suffer from the increased computational requirements of accessing particular grid cells needed in determination of the navigational goals or evaluation of the terrain traversability in planning safe and cost-efficient paths. We propose a speed-up technique to combine the benefits of uniform grid-based and tree-based representations. The proposed elevation map representation keeps the memory footprint low using tree structure but enables fast access to the grid cells corresponding to the robot surroundings. The efficiency of the proposed data representation is demonstrated in an experimental deployment of the autonomous exploration of outdoor and subterranean environments.

In this paper, we report on early results of the experimental deployment of localization techniques for a multi-rotor Micro Aerial Vehicle (MAV). In particular, we focus on deployment scenarios where the Global Navigation Satellite System (GNSS) does not provide a reliable signal, and thus it is not desirable to rely solely on the GNSS. Therefore, we consider recent advancements in the visual localization, and we employ an onboard RGB-D camera to develop a robust and reliable solution for the MAV localization in partially GNSS denied operational environments. We consider a localization method based on Kalman filter for data fusion of the vision-based localization with the signal from the GNSS. Based on the reported experimental results, the proposed solution supports the localization of the MAV for the temporarily unavailable GNSS, but also improves the position estimation provided by the incremental vision-based localization system while it can run using onboard computational resources of the small vehicle.

In this paper, we report on the deployment of the combination of commercially available off-the-shelf embedded visual localization system and RGB-D camera in an autonomous robotic exploration performed by small hexapod walking robot. Since the multi-legged walking robot is capable of traversing rough terrains, the addressed exploration problem is to create a map of an unknown environment while simultaneously performing the traversability assessment of the explored environment to efficiently and safely reach next navigational waypoints. The proposed system is targeted to run onboard of small multi-legged robots, and therefore, the system design is focused on computationally efficient approaches using relatively lightweight components. Therefore, we take advantages of the recently introduced tracking camera Intel RealSense T265 and RGB-D camera Intel RealSense D435 that are deployed to our developed autonomous hexapod walking robot that is equipped with adaptive locomotion control. Together with the proposed computationally efficient data representation and traversability assessment, the developed system supports onboard mapping and online decision-making within the exploration strategy even on a platform with low computational capabilities. Based on the reported experimental evaluation of the tracking camera, the developed system provides sufficiently accurate localization, and the robot has been able to explore indoor and outdoor environments fully autonomously.

In this paper, we address motion efficiency in autonomous robot exploration with multi-legged walking robots that can traverse rough terrains at the cost of lower efficiency and greater body vibration. We propose a robotic system for online and incremental learning of the terrain traversal cost that is immediately utilized to reason about next navigational goals in building spatial model of the robot surrounding. The traversal cost experienced by the robot is characterized by incrementally constructed Gaussian Processes using Bayesian Committee Machine. During the exploration, the robot builds the spatial terrain model, marks untraversable areas, and leverages the Gaussian Process predictive variance to decide whether to improve the spatial model or decrease the uncertainty of the terrain traversal cost. The feasibility of the proposed approach has been experimentally verified in a fully autonomous deployment with the hexapod walking robot

Avoiding collisions with obstacles and intercepting objects based on the visual perception is a vital survival ability of any animal. In this work, we propose an extension of the biologically based collision avoidance approach to the detection of intercepting objects using the Lobula Giant Movement Detector (LGMD) connected directly to the locomotion control unit based on the Central Pattern Generator (CPG) of a hexapod walking robot. The proposed extension uses Recurrent Neural Network (RNN) to map the output of the LGMD on the input of the CPG to enhance collision avoiding behavior of the robot in cluttered environments. The presented results of the experimental verification of the proposed system with a real mobile hexapod crawling robot support the feasibility of the presented approach in collision avoidance scenarios.

In this work we are concerning the problem of localization accuracy evaluation of visual-based Simultaneous Localization and Mapping (SLAM) techniques. Quantitative evaluation of the SLAM algorithm performance is usually done using the established metrics of Relative pose error and Absolute trajectory error which require a precise and reliable ground truth. Such a ground truth is usually hard to obtain, while it requires an expensive external localization system. In this work we are proposing to use the SLAM algorithm itself to construct a reliable ground-truth by offline frame-by-frame processing. The generated ground-truth is suitable for evaluation of different SLAM systems, as well as for tuning the parametrization of the on-line SLAM. The presented practical experimental results indicate the feasibility of the proposed approach.