Event
TOPS: Task-wise Adjustment of Flow Matching Source Prior and Schedule
1 reports · 1 sources
Tuo An and other authors released a paper proposing the TOPS method: without modifying the network architecture, it adjusts the flow matching source prior and interpolation schedule for each task in a language-conditioned multi-task robot manipulation flow policy. The authors report that on LIBERO-10, TOPS with single-step inference raises the success rate from 53.44% for standard flow matching with ten steps to 69.67%, with inference time about one-sixth of the baseline; on a real-robot upper drawer opening task, the success rate rises from 20% for standard flow matching to 80%. The authors also claim that the method achieves improvements on LIBERO, LIBERO-Plus, CALVIN, and in four evaluation settings when transferred to the π0.5 policy. All the above results are as reported by the paper's authors.
Report timeline
- TOPS: Reshaping Flow Source and Transport for Multi-Task Robot Manipulation
TOPS proposes a task-oriented prior and scheduling method that adapts the flow-matching source prior and interpolation schedule for each task without modifying the network architecture. On LIBERO-10, single-step inference raises the success rate from 53.44% for standard flow matching with ten steps to 69.67%, with inference time about one-sixth of the baseline. The method improves performance on LIBERO, LIBERO-Plus, and CALVIN, and when transferred to the π0.5 policy, and on a real robot it raises the success rate on an upper-drawer-opening task from 20% to 80%.
arXiv cs.RO ·