Ing. Matěj Petrlík

This paper surveys recent progress and discusses future opportunities for Simultaneous Localization And Mapping (SLAM) in extreme underground environments. SLAM in subterranean environments, from tunnels, caves, and man-made underground structures on Earth, to lava tubes on Mars, is a key enabler for a range of applications, such as planetary exploration, search and rescue, disaster response, and automated mining, among others. SLAM in underground environments has recently received substantial attention, thanks to the DARPA Subterranean (SubT) Challenge , a global robotics competition aimed at assessing and pushing the state of the art in autonomous robotic exploration and mapping in complex underground environments. This paper reports on the state of the art in underground SLAM by discussing different SLAM strategies and results across six teams that participated in the three-year-long SubT competition. In particular, the paper has four main goals. First, we review the algorithms, architectures, and systems adopted by the teams; particular emphasis is put on LIDAR-centric SLAM solutions (the go-to approach for virtually all teams in the competition), heterogeneous multi-robot operation (including both aerial and ground robots), and real-world underground operation (from the presence of obscurants to the need to handle tight computational constraints). We do not shy away from discussing the “dirty details” behind the different SLAM systems, which are often omitted from technical papers. Second, we discuss the maturity of the field by highlighting what is possible with the current SLAM systems and what we believe is within reach with some good systems engineering. Third, we outline what we believe are fundamental open problems, that are likely to require further research to break through. Finally, we provide a list of open-source SLAM implementations and datasets that have been produced during the SubT challenge and related efforts, and constitute a useful resource for researchers and practitioners.

In this paper, a novel approach for autonomously catching fast flying objects is presented, as inspired by the Mohamed Bin Zayed International Robotics Challenge (MBZIRC) 2020. In this competition, an autonomous Unmanned Aerial Vehicle (UAV) was used to intercept a ball carried by a fast flying drone. The presented solution utilizes a 3D LiDAR sensor for quick and robust target detection. The trajectory of the target is estimated and predicted to select a suitable interception position. The interceptor UAV is navigated into the interception position to safely approach the target. The interception position is frequently being adjusted based on the updated estimation and prediction of the target’s motion to ensure that the ball is caught in the dedicated onboard net. After a successful interception is detected, the UAV lands in a designated landing area. The proposed concept was intensively tested and refined in demanding outdoor conditions with strong winds and varying perception conditions to achieve the robustness required by both the demanding application and the competition. In the MBZIRC 2020 competition, our solution scored second place in Challenge 1 and first place in a combined Grand Challenge. This manuscript will provide a detailed description of the applied methods and an evaluation of our approach with data collected from real-world experiments. In addition, we present achievements of our R&D towards the transition from the MBZIRC competition to an autonomous drone interceptor, which was the main motivation of this challenge.

Integration of Visual Inertial Odometry (VIO) methods into a modular control system designed for deployment of Unmanned Aerial Vehicles (UAVs) and teams of cooperating UAVs in real-world conditions are presented in this paper. Reliability analysis and fair performance comparison of several methods integrated into a control pipeline for achieving full autonomy in real conditions is provided. Although most VIO algorithms achieve excellent localization precision and negligible drift on artificially created datasets, the aspects of reliability in non-ideal situations, robustness to degraded sensor data, and the effects of external disturbances and feedback control coupling are not well studied. These imperfections, which are inherently present in cases of real-world deployment of UAVs, negatively affect the ability of the most used VIO approaches to output a sensible pose estimation. We identify the conditions that are critical for a reliable flight under VIO localization and propose workarounds and compensations for situations in which such conditions cannot be achieved. The performance of the UAV system with integrated VIO methods is quantitatively analyzed w.r.t. RTK ground truth and the ability to provide reliable pose estimation for the feedback control is demonstrated onboard a UAV that is tracking dynamic trajectories under challenging illumination.

This paper presents a family of autonomous Unmanned Aerial Vehicles (UAVs) platforms designed for a diverse range of indoor and outdoor applications. The proposed UAV design is highly modular in terms of used actuators, sensor configurations, and even UAV frames. This allows to achieve, with minimal effort, a proper experimental setup for single, as well as, multi-robot scenarios. Presented platforms are intended to facilitate the transition from simulations, and simplified laboratory experiments, into the deployment of aerial robots into uncertain and hard-to-model real-world conditions. We present mechanical designs, electric configurations, and dynamic models of the UAVs, followed by numerous recommendations and technical details required for building such a fully autonomous UAV system for experimental verification of scientific achievements. To show strength and high variability of the proposed system, we present results of tens of completely different real-robot experiments in various environments using distinct actuator and sensory configurations.

A flexible topological representation consisting of a two-layer graph structure built on-board an Unmanned Aerial Vehicle (UAV) by continuously filling the free space of an occupancy map with intersecting spheres is proposed in this letter. Most state-of-the-art planning methods find the shortest paths while keeping the UAV at a pre-defined distance from obstacles. Planning over the proposed structure reaches this pre-defined distance only when necessary, maintaining a safer distance otherwise, while also being orders of magnitude faster than other state-of-the-art methods. Furthermore, we demonstrate how this graph representation can be converted into a lightweight shareable topological-volumetric map of the environment, which enables decentralized multi-robot cooperation. The proposed approach was successfully validated in several kilometers of real subterranean environments, such as caves, devastated industrial buildings, and in the harsh and complex setting of the final event of the DARPA SubT Challenge, which aims to mimic the conditions of real search and rescue missions as closely as possible, and where our approach achieved the 2nd place in the virtual track.

The problem of safe interception of multiple intruder UAVs by a team of cooperating autonomous aerial vehicles is addressed in this paper. The presented work is motivated by the Mohamed Bin Zayed International Robotics Challenge (MBZIRC) 2020 where this task was simplified to an interaction with a set of static and dynamic objects (balloons and a UAV), and by a real autonomous aerial interception system of Eagle.One that our team has been working on. We propose a general control, perception, and coordination system for the fast and reliable interception of targets in a 3D environment relying only on onboard sensors and processing. The proposed methods and the entire complex multi-robot sys- tem were successfully verified in demanding desert conditions, with the main focus on reliability and fast deployment. In the MBZIRC competition, the proposed approach exhibited the greatest reliability and fastest solution. It was crucial to our team in winning the entire competition and achieving the second place in the intruder UAV interception scenario.

The compact flocking of relatively localized Un- manned Aerial Vehicles (UAVs) in high obstacle density areas is discussed in this paper. The presented work tackles realistic scenarios in which the environment map is not known apriori and the use of a global localization system and communication infrastructure is difficult due to the presence of obstacles. To achieve flocking in such a constrained environment, we propose a fully decentralized, bio-inspired control law that uses only onboard sensor data for safe flocking through the environment without any communication with other agents. In the proposed approach, each UAV agent uses onboard sensors to self-localize and estimate the relative position of other agents in its local reference frame. The usability and performance of the proposed approach were verified and evaluated using various experiments in a realistic robotic simulator and a natural forest. The pre- sented experiments also validate the utility of onboard relative localization for autonomous multi-UAV applications in the ab- sence of global localization information and communication.

This paper presents a novel approach to autonomous extinguishing of indoor fires inside a building by a Micro-scale Unmanned Aerial Vehicle (MAV). In particular, controlling and estimating the MAV state, detection of a building entrance, multi-modal MAV localization during the outdoor-indoor transition, interior motion planning and exploration, fire detection and position estimation, and fire extinguishing are discussed. The performance of these elements, as well as of the entire integrated system, are evaluated in simulations and field tests in various demanding real-world conditions. The system presented here is part of a complex multi-MAV solution that won the Mohamed Bin Zayed International Robotics Challenge 2020 (MBZIRC 2020) competition, and is being used as the core of a fire-fighting Unmanned Aerial System (UAS) industrial platform under development. A video attachment to this paper is available at the website http://mrs.felk.cvut.cz/2020firechallenge-insidefires.

In this paper, a system for autonomous extinguishing of ground fires using the placement of fire blankets by Multi-rotor Unmanned Aerial Vehicles (UAVs) is proposed. The proposed system, relying on the fusion of multiple onboard sensors using only onboard computers, is infrastructure independent with a focus on high reliability in safety-critical missions that require power-on-and-go full autonomy. This task was part of the third challenge of MBZIRC 2020 aimed at the development of autonomous robotic systems for extinguishing fires inside and outside of buildings. The MBZIRC competition promotes the development of such robotics applications that are highly demanded by society and, due to their complexity and required robot abilities, go beyond the current robotic state of the art. As far as we are aware, our team was one of only two teams to achieve successful system for placement of fire blankets fully autonomously with vision-based target localization without using Real-time kinematic (RTK)-global navigation satellite system (GNSS), as was required in the competition and also for the real missions of first responders.

This letter presents a self-contained system for the robust utilization of aerial robots in the autonomous exploration of cave environments to help human explorers, first responders, and speleologists. The proposed system is generally applicable to an arbitrary exploration task within an unknown and unstructured subterranean environment and interconnects crucial robotic subsystems to provide full autonomy of the robots. Such subsystems primarily include mapping, path and trajectory planning, localization, control, and decision making. Due to the diversity, complexity, and structural uncertainty of natural cave environments, the proposed system allows for the possible use of any arbitrary exploration strategy for a single robot, as well as for a cooperating team. A multi-robot cooperation strategy that maximizes the limited flight time of each aerial robot is proposed for exploration and search & rescue scenarios where the homing of all deployed robots back to an initial location is not required. The entire system is validated in a comprehensive experimental analysis comprising of hours of flight time in a real-world cave environment, as well as by hundreds of hours within a state-of-the-art virtual testbed that was developed for the DARPA Subterranean Challenge robotic competition. Among others, experimental results include multiple real-world exploration flights traveling over 470 m on a single battery in a demanding unknown cave environment.

Autonomous operation of UAVs in a closed environment requires precise and reliable pose estimate that can stabilize the UAV without using external localization systems such as GNSS. In this work, we are concerned with estimating the pose from laser scans generated by an inexpensive and lightweight LIDAR. We propose a localization system for lightweight (under 200 g) LIDAR sensors with high reliability in arbitrary environments, where other methods fail. The general nature of the proposed method allows deployment in wide array of applications. Moreover, seamless transitioning between different kinds of environments is possible. The advantage of LIDAR localization is that it is robust to poor illumination, which is often challenging for camera-based solutions in dark indoor environments and in the case of the transition between indoor and outdoor environment. Our approach allows executing tasks in poorly-illuminated indoor locations such as historic buildings and warehouses, as well as in the tight outdoor environment, such as forest, where vision-based approaches fail due to large contrast of the scene, and where large well-equipped UAVs cannot be deployed due to the constrained space.

We present a multirotor Unmanned Aerial Vehicle (UAV) control and estimation system for supporting replicable research through realistic simulations and real-world experiments. We propose a unique multi-frame localization paradigm for estimating the states of a UAV in various frames of reference using multiple sensors simultaneously. The system enables complex missions in GNSS and GNSS-denied environments, including outdoor-indoor transitions and the execution of redundant estimators for backing up unreliable localization sources. Two feedback control designs are presented: one for precise and aggressive maneuvers, and the other for stable and smooth flight with a noisy state estimate. The proposed control and estimation pipeline are constructed without using the Euler/Tait-Bryan angle representation of orientation in 3D. Instead, we rely on rotation matrices and a novel heading-based convention to represent the one free rotational degree-of-freedom in 3D of a standard multirotor helicopter. We provide an actively maintained and well-documented open-source implementation, including realistic simulation of UAV, sensors, and localization systems. The proposed system is the product of years of applied research on multi-robot systems, aerial swarms, aerial manipulation, motion planning, and remote sensing. All our results have been supported by real-world system deployment that subsequently shaped the system into the form presented here. In addition, the system was utilized during the participation of our team from the Czech Technical University in Prague in the prestigious MBZIRC 2017 and 2020 robotics competitions, and also in the DARPA Subterranean challenge. Each time, our team was able to secure top places among the best competitors from all over the world.

In this letter we present an autonomous system intended for aerial monitoring, inspection and assistance in Search and Rescue (SAR) operations within a constrained workspace. The proposed system is designed for deployment in demanding real-world environments with extremely narrow passages only slightly wider than the aerial platform, and with limited visibility due to the absence of illumination and the presence of dust. The focus is on precise localization in an unknown environment, high robustness, safety and fast deployment without any need to install an external infrastructure such as an external computer and localization system. These are the main requirements of the targeted SAR scenarios. The performance of the proposed system was successfully evaluated in the Tunnel Circuit of the DARPA Subterranean Challenge, where the UAV cooperated with ground robots to precisely localize artifacts in a coal mine tunnel system. The challenge was unique due to the intention of the organizers to emulate the unpredictable conditions of a real SAR operation, in which there is no prior knowledge of the obstacles that will be encountered.

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.

Cooperative carrying of large, cable-suspended payloads by a pair of cooperating unmanned aerial vehicles (UAVs) is tackled in this paper. The proposed system, involving a sampling-based motion planning algorithm and a model predictive control-based coordination of UAVs, aims to achieve a smooth and reliable flight performance in environments with obstacles. The motion planning is designed to satisfy constraints on relative positions of UAVs, which are defined from the cooperative transport task and by onboard mutual localization, which is used for real-time estimation of states of neighboring robots carrying the object. Besides, a guiding principle with a cost-driven expansion is employed to steer the growth of a Rapidly-exploring Random Tree (RRT) to keep the coupled system of UAVs and the object as close to desired mutual positions as possible. A significant deviation of the controlled system from the desired configuration, by increasing or decreasing the relative distance between UAVs carrying the cable-suspended object, is achieved only if it is required by environment constraints (e.g. in narrow passages), while the allowed limits are always satisfied. Using the guiding principle enables us to find feasible solutions of the problem in a reasonable short time using onboard computer even in environments with a complicated structure of obstacles. The proposed system was evaluated in numerous simulations, compared with state-of-the-art solutions using statistical sets of results, and its performance and reliability were verified in experiments in real-world conditions.

In the task of cooperative surveillance using Micro Aerial Vehicles (MAVs), MAVs cooperatively observe a given set of Areas of Interest (AoI). The missions are usually prepared in a decoupled manner: first, the sensing locations are found, followed by computations of the trajectories assuming GPS-based localization. The precision of GPS may, however, be insufficient to keep the MAVs in compact groups, which may lead to mutual collisions. To avoid the collisions between MAVs, a camera-based on-board localization has to be used. This however requires to maintain positions of the team members in the given range to enable reliable on-board localization (each MAV has to be visible from other ones). The task of the mission planning is to find an appropriate distribution of MAVs above AoIs together with feasible trajectories from a depot to reach these locations. The on-board localization constraints and MAV motion constraints have to be satisfied during the entire mission. We propose a modification of RRT (Rapidly Exploring Random Tree) for this mission planning. The algorithm first explores the state space to find suitable sensing locations together with feasible trajectories towards them. Then, the sensing locations are optimized using Particle Swarm optimization (PSO). The proposed method has been verified in numerous simulations and outdoor experiments. The achieved results exhibit significantly better performance in terms of lower computational power and complexity of solved scenarios than the state-of-the-art solutions.

This article presents computer vision modules of a multi-unmanned aerial vehicle (UAV) system, which scored gold, silver, and bronze medals at the Mohamed bin Zayed International Robotics Challenge (MBZIRC) 2017. This autonomous system, which was running completely on-board and in real-time, had to address two complex tasks in challenging outdoor conditions. In the first task, an autonomous UAV had to find, track, and land on a human-driven car moving at $15$~$km/h$ on a figure-eight-shaped track. During the second task, a group of three UAVs had to find small colored objects in a wide area, pick them up, and deliver them into a specified drop-off zone. The computer vision modules presented here achieved computationally efficient detection, accurate localization, robust velocity estimation, and reliable future position prediction of both the colored objects and the car. These properties had to be achieved in adverse outdoor environments with changing light conditions. Lighting varied from intense direct sunlight with sharp shadows cast over the objects by the UAV itself, to reduced visibility caused by overcast to dust and sand in the air. The results presented in this paper demonstrate good performance of the modules both during testing, which took place in the harsh desert environment of the central area of United Arab Emirates, as well as during the contest, which took place at a racing complex in the urban, near-sea location of Abu Dhabi. The stability and reliability of these modules contributed to the overall result of the contest, where our multi-UAV system outperformed teams from world-leading robotic laboratories in two challenging scenarios.

This paper presents a complete pipeline for learning continuous motion control policies for a mobile robot when only a non-differentiable physics simulator of robot-terrain interactions is available. The multi-modal state estimation of the robot is also complex and difficult to simulate, so we simultaneously learn a generative model which refines simulator outputs. We propose a coarse-to-fine learning paradigm, where the coarse motion planning is alternated with guided learning and policy transfer to the real robot. The policy is jointly optimized with the generative model. We evaluate the method on a real-world platform.