Bilateral teleoperation enables robots to learn how hard to push
Compliance for Free: Learning Identifiable Impedance via Bilateral Teleoperation
Robotics
Summary
Robots know where to move but struggle to understand how hard to push during tasks that need careful contact, like wiping. The authors found a way to capture this pushing hardness, called compliance, without extra sensors by using a special teleoperation setup where a human controls the robot with two arms. This setup lets the robot learn how firm or soft to be at different times just from usual joint sensors. They tested their method on a robot wiping task and showed it adjusts force correctly when asked to wipe firmly versus normally.
What this means in practice
- •For industrial robot programmers: Program robots to perform contact tasks with varying firmness without needing extra force sensors by using bilateral teleoperation data.
- •For robotic system integrators: Integrate compliance learning into teleoperated robots to enable more natural and adaptable force control in manipulation tasks.
Authors
Harsha Guda, Adrià Colomé, Carme Torras
Abstract
Vision-language-action models tell a robot where to move, but not how hard to push. Contact-rich tasks depend on that second quantity, compliance, yet no widely used demonstration interface records it. The obstacle is identifiability as realized pose and measured force cannot separate the operator's intended equilibrium from their stiffness, so VR controllers, SpaceMouse and handheld grippers cannot supply compliance supervision even in principle. Prior compliance-output policies work around this with hand-specified task structure, privileged simulation contact state, or dedicated force and tactile hardware. Four-channel bilateral teleoperation removes the ambiguity directly by using the leader arm as a separate measurement of the intended equilibrium, making per-axis stiffness identifiable by regression using only the joint-torque sensing already on the manipulator. This yields per-timestep, direction-dependent compliance labels at zero annotation cost, which we use to fine-tune a VLA to emit stiffness alongside pose. On a Franka Research 3 wiping task, ours is the only policy of five whose contact force changes when the instruction asks for a firm wipe rather than a normal one (6.4N (normal) to 9.1N (firm) RMS, Cohen's d = 0.89, p = 0.023