VoxAct-B: Voxel-Based Acting and Stabilizing Policy for Bimanual Manipulation
This addresses sample inefficiency and limited generalization in bimanual manipulation for robotics applications, representing an incremental improvement with a novel hybrid approach.
The paper tackles the challenge of bimanual manipulation in robotics by proposing VoxAct-B, a language-conditioned, voxel-based method that uses Vision Language Models to prioritize key regions and reconstruct a voxel grid for policy learning, resulting in outperforming strong baselines on fine-grained tasks in simulation and demonstration on real-world tasks like opening drawers and jars.
Bimanual manipulation is critical to many robotics applications. In contrast to single-arm manipulation, bimanual manipulation tasks are challenging due to higher-dimensional action spaces. Prior works leverage large amounts of data and primitive actions to address this problem, but may suffer from sample inefficiency and limited generalization across various tasks. To this end, we propose VoxAct-B, a language-conditioned, voxel-based method that leverages Vision Language Models (VLMs) to prioritize key regions within the scene and reconstruct a voxel grid. We provide this voxel grid to our bimanual manipulation policy to learn acting and stabilizing actions. This approach enables more efficient policy learning from voxels and is generalizable to different tasks. In simulation, we show that VoxAct-B outperforms strong baselines on fine-grained bimanual manipulation tasks. Furthermore, we demonstrate VoxAct-B on real-world $\texttt{Open Drawer}$ and $\texttt{Open Jar}$ tasks using two UR5s. Code, data, and videos are available at https://voxact-b.github.io.