STEP: Stochastic Traversability Evaluation and Planning for Risk-Aware Off-road Navigation; Results from the DARPA Subterranean Challenge

TL;DR

This work proposes an approach to improve autonomous traversal of robots in subterranean environments that are perceptually degraded and completely unknown through a traversability and planning framework called stochastic traversability evaluation and planning (STEP).

Abstract

Although autonomy has gained widespread usage in structured and controlled environments, robotic autonomy in unknown and off-road terrain remains a difficult problem. Extreme, off-road, and unstructured environments such as undeveloped wilderness, caves, rubble, and other post-disaster sites pose unique and challenging problems for autonomous navigation. Based on our participation in the DARPA Subterranean Challenge, we propose an approach to improve autonomous traversal of robots in subterranean environments that are perceptually degraded and completely unknown through a traversability and planning framework called STEP (Stochastic Traversability Evaluation and Planning). We present 1) rapid uncertainty-aware mapping and traversability evaluation, 2) tail risk assessment using the Conditional Value-at-Risk (CVaR), 3) efficient risk and constraint-aware kinodynamic motion planning using sequential quadratic programming-based (SQP) model predictive control (MPC), 4) fast recovery behaviors to account for unexpected scenarios that may cause failure, and 5) risk-based gait adaptation for quadrupedal robots. We illustrate and validate extensive results from our experiments on wheeled and legged robotic platforms in field studies at the Valentine Cave, CA (cave environment), Kentucky Underground, KY (mine environment), and Louisville Mega Cavern, KY (final competition site for the DARPA Subterranean Challenge with tunnel, urban, and cave environments).

Figures (16)

• Figure 1: Top left: Boston Dynamics Spot quadruped robot exploring Valentine Cave at Lava Beds National Monument, CA. Top middle, bottom middle (second image from the left): Clearpath Husky robot exploring Arch Mine in Beckley, WV. Bottom left: Spot exploring abandoned Satsop power plant in Elma, WA.
• Figure 2: Overview of system architecture for STEP. From left to right: Odometry aggregates sensor inputs and relative poses. Next, Risk Map Processing merges these pointclouds and creates a multi-layer risk map. The map is used by the Geometric Path Planner and the Kinodynamic MPC Planner. An optimal trajectory is found and sent to the Tracking Controller, which produces control inputs to the robot.
• Figure 3: Estimation of ground height with uncertainty
• Figure 4: Multi-layer geometric risk analysis, which first aggregates recent pointclouds (top). Then, each type of analysis (slope, step, collision, etc.) generates a risk map along with uncertainties (middle rows). These risks are aggregated to compute the final CVaR map (bottom).
• Figure 5: Confidence-based risk analysis: the scene is illustrated through the point-of-view (POV) of a third-person (top left) and robot (top middle and top right). The aggregated pointcloud (bottom left) has regions of no returns from the area on the left side of the robot. These holes in the pointcloud are marked as negative obstacles in the risk layer (bottom middle) only when there are no returns from these regions despite them being unoccluded and sufficiently covered by laser strike pattern. This risk layer is aggregated with the geometric risk layers to compute the final CVaR map (bottom right).