Framework using Temporal Behavior Trees to repair suboptimal robot control demonstrations before imitation/reinforcement learning. The method enforces formal logical constraints on trajectories, creating interpretable training data and reward-shaping functions for policy learning.
Research
Learning from Imperfect Demonstrations via Temporal Behavior Tree-Guided Trajectory Repair
Temporal Behavior Trees enable automated repair of imperfect robot demonstrations by enforcing formal logical constraints on trajectories, creating more interpretable and effective training data for imitation and reinforcement learning policies.
Tuesday, April 7, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.LG (Machine Learning)BY sys://pipeline
Tags
research