Efficient Long-Horizon Mobile Manipulation with RAG-Assisted Vision Language Models

Le résumé fourni par la source

In dynamic and complex real-world environments, long-horizon mobile manipulation tasks impose higher demands on the autonomous capabilities of robots. Traditional methods suffer from poor generalization and high data requirements. This paper proposes a task planning method called Retrieval-Augmented generation-assisted Vision-Language Model (RAVLM). By combining retrieval-augmented generation with vision language models, RAVLM enables external knowledge retrieval and generative planning. Applied to mobile robots, this method enhances the robot’s understanding of task context and supports long-horizon task planning under complex natural language instructions. Three mobile manipulation tasks were designed on the Habitat 2.0 platform for simulation experiments. The results demonstrate that RAVLM achieves significant advantages in task planning quality, cross-task adaptability, and development cost.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.