Event

Tsinghua and National University of Singapore propose ParallelWorld embodied rehearsal planning method

1 reports · 1 sources

Researchers from Tsinghua University and the National University of Singapore propose ParallelWorld, which adds multi-step candidate future rehearsal to high-level planning for embodied agents, raising task success rate from 56% to 83%. The method generates and verifies multiple candidate action branches within a rehearsal loop; during execution it takes only the first action of the selected trajectory and then replans based on new observations. Search occurs over high-level actions with explicit execution scope and termination conditions, while low-level control is still handled by existing skills or policies. Across eight settings in LIBERO-Pro, the method's success rate is higher than the direct baseline in all of them, but adaptive search takes on average about 5.8 times as long as direct execution, and the relevant ablation covers only 3 tasks. Because the pour-liquid action could not be executed in the environment, the evaluation excluded the liquid-volume subcategory, and the remaining 28 items were included in testing.