Moving large objects, such as furniture, is a critical capability for robots operating in human environments. This task presents significant challenges due to two key factors: the need to synchronize whole-body movements to prevent collisions between the robot and the object, and the under-actuated dynamics arising from the substantial size and weight of the objects. These challenges also complicate performing these tasks via teleoperation. In this work, we introduce RobotMover, a generalizable learning framework that leverages human-object interaction demonstrations to enable robots to perform large object manipulation tasks. Central to our approach is the Dynamic Chain, a novel representation that abstracts human-object interactions so that they can be retargeted to robotic morphologies. The Dynamic Chain is a spatial descriptor connecting the human and object root position via a chain of nodes, which encode the position and velocity of different interaction keypoints. We train policies in simulation using Dynamic-Chain-based imitation rewards and domain randomization, enabling zero-shot transfer to real-world settings without fine-tuning. Our approach outperforms both learning-based methods and teleoperation baselines across six evaluation metrics when tested on three distinct object types, both in simulation and on physical hardware. Furthermore, we successfully apply the learned policies to real-world tasks, such as moving a trash cart and rearranging chairs.
We show a trained policy moving a chair towards different directions in the real world.
Same policy can generalize to other chairs and objects with varying dynamic properties.
Here we show results for a policy trained to move standing sticks(lamp and rack) and a policy trained to move tables.
We demonstrate the capability of the learned policies to address complex real-world tasks by integrating them with high-level planners. Two applications are presented: Trash Cart Transportation and Chair Rearrangement. In our experiment, we integrate a learned object manipulation policy with a teleoperation-based high-level planner, we present interactive trash cart transportation. By integrating the policy with a heuristic object root velocity planner, we showcase automatic chair rearrangement. Notably, we use the same low-level policy for both applications.
Trash Cart Transportation
Chair Rearrangement
RobotMover shows stable chair moving. Teleoperation methods cause the chair to fall off.
Comparing learning-based baselines with RobotMover, RL-EE and RL-IK result in robot self-collision and robot-object collision.
Comming Soon!