Learning Generalizable Manipulation Policy with Adapter-Based Parameter Fine-Tuning

¹ K. Lu, K. Zhou, and A. Markham are with the Department of Computer Science, University of Oxford, Oxford, UK.
² K. T. Ly, W. Hebberd, and I. Havoutis are with the Oxford Robotics Institute, University of Oxford, Oxford, UK.

            Paper Link               Code & Data (Coming Soon)

This study investigates the use of adapters in reinforcement learning for robotic skill generalization across multiple robots and tasks. We propose an alternative approach where a disembodied (virtual) hand manipulator learns a task (i.e., an abstract skill) and then transfers it to various robots with different kinematic constraints without retraining the entire model (i.e., the concrete, physical implementation of the skill). Example results in simulation & real (click).

We present generalizing a learned skill i.e., opening a drawer or a cupboard, to different robotic platforms, including HSR, AliengoZ1, and A2Single.

We present generalizing a learned skill to different tasks, such as from drawer opening to door opening and chair pushing.

In the real experiments, we perform the drawer opening task and chair pushing task with a real HSR robot.

The below video shows the chair pushing task.

Have any questions, please feel free to contact Kai Lu