PolyBot: Training One Policy Across Robots While Embracing Variability

Jonathan Yang, Dorsa Sadigh, Chelsea Finn

jyang27@cs.stanford.edu

Motivation

Method

Our framework aligns the input, output, and internal representation spaces of our policy across embodiments.

Contrastive Internal Representation Alignment

Our Robots!

We evaluate our method on the Franka Emika Panda, Rethink Robotics Sawyer, and Trossen Robotics WidowX 250S. Each of these robots have differing sizes and kinematic configurations.

Evaluation

Distractor Pick/Place

New Object Pick/Place

New Container Pick/Place

Reversed Container Shelf Manipulation

New Compartment Shelf Manipulation

Paper