Event

Continuous-Time Safety Planning Framework for Quadruped Robots

1 reports · 1 sources

Gage MacLin and other authors propose a real-time motion planning framework for agile quadruped robots that retains the robot's full SE(2) model and its anisotropic, heading-coupled velocity constraints, and accepts the resulting non-convexity. The method parameterizes trajectories as composite Bernstein polynomials, uses the convex hull property to prove obstacle avoidance and input constraints hold over the entire planning horizon, and runs in a closed loop at an online replanning frequency, sending body velocities to the motion controller. The authors report that in simulation they compare against ablation methods and external baselines, and validate in hardware experiments where a quadruped robot navigates among unknown moving obstacles, recovering lateral and heading-coupled motions that simplified models exclude.