Clarification on Dynamic Action Library Updates in ROS-LLM

[yan12125]

I learned about ROS-LLM at IROS 2025. Thanks for the great presentation — the ROS-LLM framework is impressive!

I’m particularly interested in the dynamic action library updates via imitation learning mentioned in Section 3.6 of the arXiv paper (https://arxiv.org/abs/2406.19741). I’ve looked through the codebase (https://github.com/huawei-noah/HEBO/tree/rosllm/ROSLLM) but couldn’t find the relevant implementation.

Could you point me to the related code or share any guidance? Thank you!