Stageguard improves robot skill switching for long tasks
StageGuard: Learning Stage Transitions for Long-Horizon Robot Tasks via Agentic Distillation
Robotics
Summary
Robots often need to perform a series of smaller actions to complete a big task, and knowing exactly when to switch between these smaller actions is hard. The authors developed a system called StageGuard that teaches a smaller, faster model to recognize when one step is done and the next should start. This approach uses detailed explanations generated by a larger model combined with real demonstrations to train the smaller model efficiently. Their method improves accuracy in predicting task stage changes and works well in real-time robot control.
What this means in practice
- •For robotics engineers: Enhance robot controllers to more reliably detect subtask completions for smooth multi-step task execution.
- •For automation system integrators: Integrate efficient stage-transition detection to monitor long robotic tasks with less delay and higher accuracy.
Authors
Jinbang Huang, Yuanzhao Hu, Zhiyuan Li, Ran Qi, Yixin Xiao, Yangzheng Wu, Tengyue Ba, Zhanguang Zhang, Yingxue Zhang
Abstract
Hierarchical planning frameworks combine skills from multiple robot control policies for long-horizon task execution, where determining when to terminate the current skill and advance to the next subtask is essential. Existing approaches often rely on pre-designed completion signal checkers that are hard to obtain in real-world execution. Large-scale vision-language models (VLMs) offer strong reasoning capabilities, but their decision boundaries are not inherently aligned with task completion criteria, while cloud deployment and lengthy reasoning introduce substantial latency, limiting real-time monitoring. We propose StageGuard, an agentic distillation framework for accurate and efficient stage-transition decisions. StageGuard combines teacher-model reasoning with demonstration trajectories to generate structured explanations of subtask completion and policy switching. A lightweight student VLM uses these explanations to generate compact self-explanations, which are used for supervised fine-tuning. We evaluate stage-transition prediction on trajectories from two benchmarks and assess closed-loop task success through integration into hierarchical robot control on BEHAVIOR-1K, with further validation on real robots. Results show substantial improvements in stage-transition prediction while supporting efficient online monitoring.