ArtPlanner: Robust Legged Robot Navigation in the Field

Abstract

Due to the highly complex environment present during the DARPA Subterranean Challenge, all six funded teams relied on legged robots as part of their robotic team. Their unique locomotion skills of being able to step over obstacles require special consideration for navigation planning. In this work, we present and examine ArtPlanner, the navigation planner used by team CERBERUS during the Finals. It is based on a sampling-based method that determines valid poses with a reachability abstraction and uses learned foothold scores to restrict areas considered safe for stepping. The resulting planning graph is assigned learned motion costs by a neural network trained in simulation to minimize traversal time and limit the risk of failure. Our method1 achieves real-time performance with a bounded computation time. We present extensive experimental results gathered during the Finals event of the DARPA Subterranean Challenge, where this method contributed to team CERBERUS winning the competition. It powered navigation of four ANYmal quadrupeds for 90 minutes of autonomous operation without a single planning or locomotion failure.

Locomotion Policy Guided Traversability Learning using Volumetric Representations of Complex Environments

“…Model-based approaches for assessing geometric traversability have attracted continuous research, particularly in the context of wheeled robots, where the wheel-to-terrain interactions can be more easily defined compared to more complex legged systems (Wellhausen and Hutter, 2023). These approaches demonstrate good generalization, specifically when the underlying assumptions like the rigidity of the environment, hold.…”

“…These approaches demonstrate good generalization, specifically when the underlying assumptions like the rigidity of the environment, hold. For this, recent works rely on analysis of point cloud data (Fan et al, 2021;Cao et al, 2022;Xue et al, 2023a), mesh data (Hudson et al, 2022), or elevation maps (Bouman et al, 2020;Fan et al, 2021;Wellhausen and Hutter, 2023). On the other hand, data-driven approaches use simulation to collect data in a trial-and-error fashion to estimate traversability for wheeled (Guzzi et al, 2020) or legged robots (Frey et al, 2022).…”