SuperNav: An Agentic Navigation System for Any Task in Any Scene
SuperNav uses an unfine-tuned MLLM plus navigation tools for general robot navigation.
SuperNav is an agentic navigation system that uses a pretrained multimodal LLM to interpret requests, understand scenes, and decide actions without navigation-specific fine-tuning. Motion is delegated to Navigation Skills and physical-interaction tools, with a visual-point interface that lets the model mark destinations in images and revise plans from execution feedback. It outperforms four baselines on instance-level, multi-object, and demand-driven tasks, with category-level results on HM3D and deployment on a real quadruped.
- SuperNav keeps a pretrained MLLM for decisions and delegates motion to tools.
- No navigation-specific fine-tuning of the MLLM.
- A visual-point interface lets the model set destinations in images and revise from feedback.
- It beats four baselines and was deployed on a real quadruped.
From the original article
1 min read at huggingface.co
General-purpose service robots need navigation systems that can handle diverse human requests in unfamiliar environments, combining task generality with scene generality.
By Jinkai Zhang, Jingyi Xu, Yuanhong Yu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.