3

ActionFlow: Equivariant, Accurate, and Efficient Policies with Spatially Symmetric Flow Matching
Best Paper Award in the RSS 2024 Workshop on Priors4Robots This paper introduces a novel policy class combining Flow Matching with SE(3) Invariant Transformers for efficient, equivariant, and expressive robot learning from demonstrations. We showcase the performance of the method across simulation and real robot environment considering low dimensional as well as high dimensional real-world observations.
Niklas Funk, Julen Urain, Joao Carvalho, Vignesh Prasad, Georgia Chalvatzaki, Jan Peters
PDF Cite Project
ActionFlow: Equivariant, Accurate, and Efficient Policies with Spatially Symmetric Flow Matching