RoLoMa

Real-world deployment demands robustness to model mismatch, sensor noise, and communication delays. Controllers can react to those at run time — but online execution can only ever be as robust as the motion plan allows. RoLoMa makes robustness a first-class objective at the planning stage.

We derive a metric from first principles that represents robustness against external disturbances — the smallest unrejectable force — and use it inside our trajectory optimisation framework to solve complex loco-manipulation tasks on an ANYmal quadruped with a robotic arm.

Results

Trajectories generated with this approach resist a greater range of forces, originating from any direction. They solve the tasks as effectively as before, with the added benefit of counteracting stronger disturbances in worst-case scenarios — demonstrated on hardware turning a hand wheel, pulling a lever, and lifting increasingly heavy loads.

Published in Autonomous Robots (2023). The paper page has the full abstract and all the supplementary videos.