HeckNav: Interpretable Heuristic Knowledge Navigation via Bayesian Probabilistic Inference

Xianshun Jiang, Yao Xiong, Yanjia Huang, Xiangbo Gao, Zhengzhong Tu

Keywords: Embodied AI, Explainability, Knowledge Graph, Object Navigation

Abstract: Object goal navigation (ObjectNav) is a well defined and fundamental problem for robotic to navigate, that starting in a random point at unseen environment to a given object. Recent advanced study in ObjectNav uses Vision-Language Models (VLMs) or multimodal large language models(MLLMs) for object detection and decision-making. However, due to the ”black-box” nature and implicit reasoning of VLMs, current ObjectNav systems often rely on the opaque internal capabilities of the models or exhaustive prompt engineering, lacking transparency and explicit planning. In this paper, we propose HeckNav, a zero-shot, heuristic-based knowledge navigation module that enhance VLMs with Knowledge Graph (KG) and Bayesian probabilistic inference. To enhance planning efficiency while maintaining a minimal footprint, we introduce a lightweight KG module that employ Bayesian inference to derive navigational plans based on the robot’s observed environment. Furthermore, our framework provides unique level of interpretability by visualizing the agent’s environmental awareness and its path-finding process through an abstract graph based on heuristic value preferences. Experimental results demonstrate that HeckNav achieves significant improvement on the HM3D and MP3D datasets especially on smaller VLM

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 2