Learning Deep Parameterized Skills from Demonstration for Re-targetable Visuomotor Control
This addresses the challenge of re-targetable visuomotor control for robots, enabling zero-shot generalization to new goals, which is incremental over prior methods that required training per goal or lacked target inference from visual data.
The paper tackles the problem of robots learning skills that can generalize to unseen goals by introducing an end-to-end method using a goal-parameterized neural network policy, achieving over 90% generalization to targets in simulation and robot experiments when trained on only 33% of possible goals.
Robots need to learn skills that can not only generalize across similar problems but also be directed to a specific goal. Previous methods either train a new skill for every different goal or do not infer the specific target in the presence of multiple goals from visual data. We introduce an end-to-end method that represents targetable visuomotor skills as a goal-parameterized neural network policy. By training on an informative subset of available goals with the associated target parameters, we are able to learn a policy that can zero-shot generalize to previously unseen goals. We evaluate our method in a representative 2D simulation of a button-grid and on both button-pressing and peg-insertion tasks on two different physical arms. We demonstrate that our model trained on 33% of the possible goals is able to generalize to more than 90% of the targets in the scene for both simulation and robot experiments. We also successfully learn a mapping from target pixel coordinates to a robot policy to complete a specified goal.