Sparse Compositional Flow Matching for Embodied AI Trajectories
A new generative model, Sparse Compositional Flow Matching, is proposed for embodied AI trajectories. The method addresses sample inefficiency by modeling trajectories as compositions of reusable motion primitives rather than dense monolithic signals. It composes directly in physical space, avoiding post-hoc decoding. The approach targets robotic manipulators, underwater vehicles, and mobile robots. The paper is available on arXiv under ID 2605.23341.
Key facts
- arXiv ID: 2605.23341
- Announce Type: cross
- Method: Sparse Compositional Flow Matching
- Composes directly in physical space
- Uses reusable motion primitives
- Targets robotic manipulators, underwater vehicles, mobile robots
- Addresses sample inefficiency in generative models
- Published on arXiv
Entities
Institutions
- arXiv