Learning Force Control for Legged Manipulation

Tifanny Portela     Gabriel B. Margolis     Yandong Ji     Pulkit Agrawal

Controlling the contact force during interactions is an inherent requirement for locomotion and manipulation tasks. Current reinforcement learning approaches to locomotion and manipulation rely implicitly on forceful interaction to accomplish tasks but do not explicitly regulate it. This paper proposes a reinforcement learning task specification that focuses on matching desired contact force levels. Integrating force control with the coordination of a robot’s body and arm, we present an end-to-end policy for legged manipulator control. Force control enables us to realize compliant gripper and whole-body pulling movements that have not been previously demonstrated using a learned policy. It also facilitates a characterization of the force-tracking performance of learned policies in simulation and the real world, indicating their performance potential for force-critical tasks.

We train a reinforcement learning controller to track force commands in simulation and transfer to a full-sized quadruped manipulator.

We also implement a locomotion and manipulation controller to complete a pipeline for teleoperation.

By adjusting the force command input to the policy, we can compensate for gravity to support a payload while maintaining compliance.

One use case for the compliant policy is kinesthetic demonstration on a large, high-degrees-of-freedom robot manipulating heavy objects.