Rethink Before You Execute: Adaptive Execution for World Action Models
2026-08-10 • Robotics
Robotics
AI summaryⓘ
The authors study World Action Models (WAMs), which predict actions and environment changes for robots, but note that using a fixed time to replan actions is not always efficient. They propose TempoWAM, a method that watches how well the robot is progressing during execution and decides when to replan based on actual progress rather than a fixed step count. Their approach adapts over time to better fit different tasks and leads to better performance and fewer unnecessary replans in both simulations and real robots. TempoWAM helps robots complete tasks more efficiently and successfully according to the authors.
World Action Modelsrobot replanningtask progress monitoringadaptive executionrecurrent progress monitorexecution dynamicsinference efficiencyrobotic task planning
Authors
Feng Ye, Yiming Zhao, Yong Yu, Hongxu Zhou, Yong Pan, Yuan Xue, Peng Jia, Chuanmin Jia
Abstract
World Action Models (WAMs) jointly predict future actions and the evolution of the environment. At each inference, a WAM generates a chunk of actions and the robot executes a fixed prefix before replanning. We argue that this fixed execution horizon is poorly matched to execution dynamics: the chunk reliability varies across task stages, so when to replan depends on the result of accumulated execution, not on the step counts. We propose TempoWAM (Timing Execution by Monitoring Progress Online), a lightweight plug-and-play execution scheme for WAMs. A Recurrent Progress Monitor first estimates task progress from the current observation, task instruction, remaining actions, and execution history; and an Adaptive Execution Protocol then evaluates whether the chunk is advancing the task to decide if replanning is needed. To bridge the training-deployment gap, the protocol is calibrated by a task-dependent calibration factor with online adaptation. Experiments on LIBERO, RoboTwin, and real-world tasks show that TempoWAM consistently improves the efficiency-success trade-off of WAM execution. On real robots, it reduces WAM inferences by 26.9% on easy tasks while maintaining success, and improves success by 13.3 points on difficult tasks.