Teleoperating a humanoid robot becomes harder when the robot picks up a tool or load, because the robot's dynamics change in ways that are hard to predict. A new paper on arXiv proposes a control system that estimates these changes in real time and adjusts the robot's whole-body motion accordingly.

The approach combines whole-body model predictive control (MPC) with online parameter estimation. The MPC plans motions across the entire body, while the estimator tracks shifts in inertial properties as they happen. This lets the operator maintain stable control even when the payload is significant.

The authors argue that such adaptation is essential for practical deployment of humanoids in tasks like heavy tool use or carrying loads. By estimating inertial parameters online, the system reduces the need for manual tuning or pre-known payload specifications, a common limitation in current teleoperation setups.