2nd Workshop on Informative Path Planning and Adaptive Sampling

Abstract: The measurement of environmental parameters, either from space, in-situ systems, or by a combination of both, has never been so intense, fostered by climate change and the pressing need to ensure Earth's environmental sustainability. The already massive flow of data will escalate with the increasing adoption of low cost remote sensing solutions, the miniaturization of sensors, and the deployment of autonomous platforms. However, to improve understanding on the Earth system and its interacting components, one needs not only more data and to ensure its automatic processing, but eventually "less and better" data and the ability to extract meaningful information from it. Despite many advances in earth data science, the typical variability of earth observations across multiple and interacting spatial and temporal scales remains challenging to describe and interpret. Here I review basic fundamental concepts for the extraction of meaningful information from temporal and spatial data from an applied earth science perspective.

Speaker bio: Susana Barbosa is a senior researcher at INESC TEC (Porto, Portugal) working in the Centre for Information Systems and Computer Graphics (CSIG) and in the Centre for Robotics and Autonomous Systems (CRAS), at the interface of data science, earth observation, and robotics. Her research is highly interdisciplinary, with a strong emphasis on climate and earth system science. Her current research interests include the application of autonomous systems and robotic platforms for monitoring extreme environments and the study of space-atmosphere-ocean interactions.

Abstract: A fundamental task of robotic systems is to use on-board sensors and perception algorithms to understand high-level semantic properties of an environment. These semantic properties may include a map of the environment, the presence, pose and class of objects, the behaviour of other agents, or the parameters of a dynamic field. The performance of perception algorithms can be greatly improved by planning the motion of the robots to obtain high-value observations. This motivates the problem of active perception, where the goal is to plan the observation viewpoints for a team of robots while considering both the motion constraints and the perception objectives of the task at hand. This talk will present the decentralised Monte Carlo tree search (Dec-MCTS) algorithm, which generalises standard MCTS for multi-robot active perception. An extended discussion will also be provided regarding the formulation of various objective functions suitable for applications including marine environmental monitoring and precision agriculture.

Speaker bio: Graeme Best is a Postdoctoral Scholar in the Robotic Decision Making Laboratory (RDML) at Oregon State University (OSU), advised by Prof. Geoffrey Hollinger. He is currently a team member in the joint Carnegie Mellon University and OSU team for the DARPA Subterranean Challenge, and is also involved with marine environmental monitoring research projects. He received his Ph.D. at the Australian Centre for Field Robotics (ACFR) at the University of Sydney (2019), and the B.E. (Hons) (Electrical) and B.Sc. (Computer Science) from Monash University (2014).

Abstract: Having a set of locations where sensor measurements can be retrieved, the problem of finding a cost-efficient path to collect data from the locations can be addressed as one of the existing variants of the routing problems such the Traveling Salesman Problem (TSP) or Orienteering Problem (OP). However, a robotic system might be constrained by its motion capabilities, and therefore, there is a fundamental requirement on the feasibility of the found solutions. In this talk, I will present an overview of existing approaches for curvature-constrained systems that can be modeled using Dubins vehicle together with solution quality estimation based on tight lower bounds for Dubins TSP. Besides, a generalization of the data collections planning in 3D, especially suitable for multi-rotor vehicles, will be overviewed. Moreover, recent results on a combination of the routing with motion planning will be introduced as the Physical Orienteering Problem (POP) that provides a general way how to address data collection planning with vehicles for which feasible trajectories can be found by randomized sampling-based motion planning methods.

Speaker bio: Jan Faigl is an associate professor of computer science at the Department of Computer Science, Faculty of Electrical Engineering (FEE), Czech Technical University in Prague (CTU), Czechia. He received the Ph.D. degree in artificial intelligence and biocybernetics and the Ing. degree in cybernetics from CTU in 2010 and 2003, respectively. Since 2013, he leads Computational Robotics Laboratory within the Artificial Intelligence Center (AIC). He is also co-founder of the Center for Robotics and Autonomous Systems (CRAS), where he is currently participating on DARPA Subterranean Challenge. Since 2019, he is serving as editor of IEEE Transactions on Automation Science and Engineering. His current research interests include robotic information gathering, multi-goal and data collection planning, autonomous navigation, path and motion planning techniques, incremental learning, robotic learning, and locomotion control of multi-legged walking robots.

Abstract: Many mobile robots today are capable of collecting information in the form of sensor readings. However, what sensor readings lead to "desirable" information and how to collect them are both ambiguous and challenging questions; such information could be on parameters of an unknown vector field, or hidden intent of traffic participants on a public road. Information-theoretic active sensing, in which a robot closes the loop from perception to action in search of optimal control under an information-theoretic cost, sheds light on those questions from an optimization perspective. The key components in the information-theoretic active sensing is the online inference of partially observable stochastic processes and the effective planning to reduce their uncertainty. In this talk, I will summarize recent and ongoing projects on this topic within the Multi-Robot Systems Lab at Stanford University. The algorithmic strategies range from RRT path planning to Belief MDP, and the applications include environmental monitoring, target tracking, and intent inference.

Speaker bio: Haruki Nishimura is a Ph.D. Candidate in Aeronautics and Astronautics at Stanford University. He completed his M.S. in Aeronautics and Astronautics at Stanford University in 2017, and received his B.Eng. in Aeronautics and Astronautics from the University of Tokyo in Japan in 2015. Haruki's current research interests include belief-space planning, motion-based communication, and informative motion planning.

Abstract: Problems in optimal planning for marine vehicle systems are discussed, formulated, and solved in the framework of dynamic programming techniques. This is done with reference to the application of the Principle of Optimality and to the development of numerical methods used to solve the Hamilton-Jacobi Bellman equation. The focus is on problems with logic-based constraints for single-vehicle operation and coordination constraints for multi-vehicle operations. This is done with reference to Y. C. Ho’s generalized control framework—in which there is more than one criterion and more than one intelligent controller, each of which having access to different information—to show that we may have been missing something in coordination and control since the 70’s. A few numerical examples are discussed to illustrate the main concepts, as well as problems associated to the application of the Principle of Optimality.

Speaker bio: João Tasso de Figueiredo Borges de Sousa is with the Electrical and Computer Engineering Department from Porto University in Portugal. He holds a PhD and an MSc in Electrical Engineering, both awarded by Porto University. His research interests include autonomous underwater, surface and air vehicles, planning and execution control for networked vehicle systems, optimization and control, cyber-physical systems, and applications of networked vehicle systems to the ocean sciences, security and defense.

He is the head of the Laboratório de Sistemas e Tecnologias Subaquáticas – LSTS (Underwater Systems and Technologies Laboratory). The LSTS developed the award-winning Light Autonomous Underwater Vehicle (LAUV) and the LSTS open source software tool chain for networked vehicle systems (https://www.lsts.pt/toolchain).

He has been organizing the annual Rapid Environmental Picture MUS exercise in cooperation with the Portuguese Navy since 2010, and with the Centre for Maritime Research and Experimentation since 2014. He was the chair of the 2013 edition of the IFAC Navigation, Guidance and Control Workshop and of the 2018 IEEE AUV Symposium. He is a member of the Advisory Board of the Swedish Marine Robotics Center. He is in the editorial board of several scientific journals. He is a member of several NATO committees. He has authored over 400 publications, including 40 journal papers.

Abstract: Efficient robotic exploration has the potential to solve significant challenges and save lives when disasters occur underground. Cave rescues are demanding endeavors performed in unmapped environments with limited communications due to the convoluted nature of underground voids. Nuclear waste storage facilities become inaccessible to humans when radiation leaks occur, so efficient pose estimation and mapping to localize radiation leaks is of the utmost importance.

Without reliable communications, robots must operate autonomously to explore these environments, but many Simultaneous Localization and Mapping (SLAM) techniques do not generate maps fit for active perception. Occupancy grid maps, which are used to generate feasible trajectories, may be updated quickly, but the gains in speed come at the cost of memory efficiency and fidelity. For computationally constrained systems, employing low-resolution environment representations may obscure small passageways and hazards or obliterate rich details.

Gaussian Mixture Models (GMMs) are well suited to compactly represent sensor observations and model the structural correlations present in the environment. These generative models are advantageous as compared to voxelized representations that assume independence between cells and lose dependencies between spatially distinct locations. GMM-based perception tasks such as registration have been studied, but solutions are either not real-time viable or have not been evaluated with large, real-world datasets. There are few works that address the viability of leveraging these models for tasks such as SLAM and exploration, because a significant challenge to overcome is the time needed to create these models.

This talk will present information-theoretic exploration with approximate continuous distributions that unifies high-resolution, low memory footprint environment modeling with occupancy mapping techniques while remaining amenable to local and global pose updates. This is possible through innovations in robust distribution to distribution registration, arbitrary resolution occupancy modeling, real-time information-theoretic exploration, and simultaneous localization and mapping. These developments are evaluated in simulation, with real-world datasets, and onboard an aerial system and tested in complex environments. The results demonstrate that leveraging compact generative models yields substantial gains over state-of-the-art methods in model fidelity, accuracy, and memory efficiency.

Speaker bio: Wennie is a Postdoctoral Fellow in the Robotics Institute at Carnegie Mellon University. She researches perception, planning, and learning algorithms to enable safe autonomy in significantly three-dimensional, complex environments. Her current research develops methods to enable aerial systems to explore subterranean environments.