Talking Robotics: Active Perception and Exploration with Teams of Robots — From Simulation to Subterranean
This seminar talk presents most of the work from my thesis along with a brief discussion of my time at JPL competing in the DARPA Subterranean Challenge. I discuss topics such as mutual information objectives for exploration, algorithms for distributed perception planning in multi-robot teams, and scalable planning for target tracking problems.
Many thanks to the Talking Robotics team (Patrícia Alves-Oliveira, Silvia Tulli, and Miguel Vasco) for organizing!
IROS 2021: Scalable Distributed Planning for Multi-Robot, Multi-Target Tracking
This work applies Randomized Sequential Partitions (RSP) to multi-robot, multi-target tracking with a mutual information objective. RSP is a distributed algorithm for sensor and perception planning and an approximation algorithm for submodular maximization problems. The analysis in this work develops new bounds for mutual information objectives that form a sum over targets, and the results focus on scaling to large teams and demonstrate receding-horizon planning for a team of 96 robots.
ICRA 2021: Volumetric Objectives for Multi-Robot Exploration of Three-Dimensional Environments
This work contrasts coverage-based and information-theoretic perception objectives for exploration (autonomous mapping) and attempts to unify these two bodies of work. Our analysis indicates that coverage objectives can be good approximations for information-theoretic quantities, and initial simulation results indicate that coverage-based objectives can provide significantly better exploration performance than existing beam-based mutual information objectives, particularly when accounting for multiple observations.