Scaling Manipulation Learning with Visual Kinematic Chain Prediction
This addresses the challenge of scaling robot manipulation learning across different robots and workspaces without manual adjustments.
The paper tackles the problem of learning general-purpose robot manipulation models across diverse environments by introducing visual kinematic chains as a universal representation that eliminates manual action normalization. The proposed Visual Kinematics Transformer (VKT) achieves superior performance over baseline methods on multiple benchmarks including Calvin, RLBench, Open-X, and real robot tasks.
Learning general-purpose models from diverse datasets has achieved great success in machine learning. In robotics, however, existing methods in multi-task learning are typically constrained to a single robot and workspace, while recent work such as RT-X requires a non-trivial action normalization procedure to manually bridge the gap between different action spaces in diverse environments. In this paper, we propose the visual kinematics chain as a precise and universal representation of quasi-static actions for robot learning over diverse environments, which requires no manual adjustment since the visual kinematic chains can be automatically obtained from the robot's model and camera parameters. We propose the Visual Kinematics Transformer (VKT), a convolution-free architecture that supports an arbitrary number of camera viewpoints, and that is trained with a single objective of forecasting kinematic structures through optimal point-set matching. We demonstrate the superior performance of VKT over BC transformers as a general agent on Calvin, RLBench, Open-X, and real robot manipulation tasks. Video demonstrations can be found at https://mlzxy.github.io/visual-kinetic-chain.