Deductive Chain-of-Thought Augmented Socially-aware Robot Navigation World Model

TL;DR

The paper addresses unreliable, ungrounded LLM-driven social robot navigation by introducing NaviWM, which integrates a grounded spatial-temporal world model with a deductive chain-of-thought reasoning framework. The world model represents agents and interactions as a structured graph and feeds formal, first-order logic constraints into the LLM, while a Gentzen-tree-based deductive procedure governs multi-step reasoning and verification. Key Contributions include (1) a world model for social navigation, (2) a formal deductive CoT framework, and (3) an integrated system that provides verifiable, socially-aware plans with improved safety in dynamic human spaces. Empirical results show NaviWM outperforms baselines across success rate and social-violation metrics, achieving up to SR ≈ 0.8 in moderate crowds and maintaining safe, efficient behavior in dense crowds, along with substantial reductions in uncomfortable interactions.

Abstract

Social robot navigation increasingly relies on large language models for reasoning, path planning, and enabling movement in dynamic human spaces. However, relying solely on LLMs for planning often leads to unpredictable and unsafe behaviors, especially in dynamic human spaces, due to limited physical grounding and weak logical consistency. In this work, we introduce NaviWM, a socially-aware robot Navigation World Model that augments LLM reasoning with a structured world model and a logic-driven chain-of-thought process. NaviWM consists of two main components: (1) a spatial-temporal world model that captures the positions, velocities, and activities of agents in the environment, and (2) a deductive reasoning module that guides LLMs through a multi-step, logic-based inference process. This integration enables the robot to generate navigation decisions that are both socially compliant and physically safe, under well-defined constraints such as personal space, collision avoidance, and timing. Unlike previous methods based on prompting or fine-tuning, NaviWM encodes social norms as first-order logic, enabling interpretable and verifiable reasoning. Experiments show that NaviWM improves success rates and reduces social violations, particularly in crowded environments. These results demonstrate the benefit of combining formal reasoning with LLMs for robust social navigation. Additional experimental details and demo videos for this work can be found at: https://sites.google.com/view/NaviWM.

Figures (3)

• Figure 1: The world model illustration of social robot navigation scenario: NaviWM constructs the world model from local observation to capture both agent vertex information and environmental semantic information with respect to spatial-temporal HRI features.
• Figure 2: The architecture of NaviWM: (1). NaviWM constructs a world model to represent environmental description from local observation; (2). The deductive CoT algorithm is encoded as prompt engineering for LLM reasoning guidance; (3). The inference chains are generated step-by-step with respect to logical procedure and self-validation step; (4). Final robot action is obtained in the final step of the deductive CoT.
• Figure 3: Verification proof tree in NaviWM illustrating the deductive reasoning process that derives the collision-free action $a_0$ from the objective event $\Phi(a)$.