Experience-guided search improves skill use in new robot tasks
Experience-Guided Initiation Search for Learned Skills in Skill Composition
Robotics
Summary
Robots often learn specific skills that work well in certain situations, but using those skills in new places can be tricky because they need the right starting conditions. The authors propose a way to find these good starting points by combining past experiences with checks in the new environment without needing lots of trial runs. Their method helps robots more efficiently figure out when and how to use learned skills in new tasks, especially when multiple skills are combined. This reduces the time and effort needed to get the robot working reliably in unfamiliar settings.
What this means in practice
- •For robotics engineers: Optimize robot task execution in new settings by efficiently identifying reliable starting conditions for pre-trained skills using combined experience and environment checks.
- •For automation integrators: Reduce setup times for robotic systems performing multi-step manipulation tasks by guiding skill initiation choices with past data and minimal environment trials.
Authors
Qixuan Li, Yanhong Zhao, Jincheng Yu
Abstract
Deploying frozen learned skills, such as Vision-Language-Action (VLA) policies, in new environments requires identifying initiation configurations that support reliable execution. In skill composition, an initiation configuration affects not only the current skill but also the physical state passed to subsequent skills, so successful execution of an individual skill does not necessarily imply successful completion of the composed task. Estimating target-specific capability through extensive rollouts is costly in real-world deployment, while directly reusing historical experience can be unreliable under environment changes. We propose EVIS, an Experience-Guided and Behavior-Validated Initiation Search framework for discovering reliable initiation configurations under limited target interaction. EVIS uses historical execution experience to prioritize promising candidates and target-environment behavior to validate whether they remain effective. We evaluate EVIS on single-skill and two-stage manipulation tasks with frozen VLA policies. EVIS reduces mean target-environment queries and improves reliable candidate discovery under small interaction budgets. These results show that combining historical guidance with target-side behavioral validation can reduce the interaction cost of deploying frozen learned skills in new environments.