Article

DMAT: Direct Action-Manifold Modeling for Efficient and Robust Bimanual Manipulation

The authors propose the Direct Manifold Action Transformer (DMAT), which reframes bimanual action generation as directly modeling clean, observation-conditioned joint-space action chunks from stochastic interpolants, enabling single-step inference without iterative denoising or trajectory straightening.

Part of the event: DMAT: Single-Step Inference Directly Models Bimanual Action Manifolds

This page provides a summary and source link only.