GameChat: Multi-LLM Dialogue for Safe, Agile, and Socially Optimal Multi-Agent Navigation in Constrained Environments

TL;DR

GameChat introduces a decentralized framework for safe, agile, and socially optimal multi-agent navigation in constrained environments by leveraging natural language dialogue between agents to resolve conflicts and break symmetry. It combines an LLM-based priority negotiation module with a subgame-perfect equilibrium controller and control barrier functions for safety, achieving significant improvements over baselines in doorways and intersections. The paper provides a formal POSG formulation, details the methodology, and presents extensive simulations showing substantial reductions in makespan and perfect priority alignment. The work advances human-robot coexistence by enabling robust, decentralized coordination without central authority.

Abstract

Safe, agile, and socially compliant multi-robot navigation in cluttered and constrained environments remains a critical challenge. This is especially difficult with self-interested agents in decentralized settings, where there is no central authority to resolve conflicts induced by spatial symmetry. We address this challenge by proposing a novel approach, GameChat, which facilitates safe, agile, and deadlock-free navigation for both cooperative and self-interested agents. Key to our approach is the use of natural language communication to resolve conflicts, enabling agents to prioritize more urgent tasks and break spatial symmetry in a socially optimal manner. Our algorithm ensures subgame perfect equilibrium, preventing agents from deviating from agreed-upon behaviors and supporting cooperation. Furthermore, we guarantee safety through control barrier functions and preserve agility by minimizing disruptions to agents' planned trajectories. We evaluate GameChat in simulated environments with doorways and intersections. The results show that even in the worst case, GameChat reduces the time for all agents to reach their goals by over 35% from a naive baseline and by over 20% from SMG-CBF in the intersection scenario, while doubling the rate of ensuring the agent with a higher priority task reaches the goal first, from 50% (equivalent to random chance) to a 100% perfect performance at maximizing social welfare.

Figures (8)

• Figure 1: Two agents head toward a hospital and a grocery store in a symmetric, constrained environment. Both agents need to get through the gap, but one must go in front or they will collide. In the left image, there is no communication between the agents, causing a deadlock as the agents do not know which should go first, and in the right image, GameChat uses natural language communication between decentralized agents to identify their roles, thereby resolving the deadlock by prioritizing urgent tasks.
• Figure 2: Example of a social mini-game. Each agent (represented by the red and blue rectangles) has length $l$ and must pass through $\mathcal{Q}$ on the way to their goals, $\mathcal{X}^1_g$ and $\mathcal{X}^2_g$. The desired trajectories $(\widetilde{\Gamma}^1, \widetilde{\Gamma}^2)$ for each agent are represented by the arrows. Note that they intersect at $\mathcal{Q}$ at $t=2$.
• Figure 3: Flow chart describing the logical flow of GameChat. Blue box represents nodes involved in our novel LLM-based communication module. Orange box represents nodes handling a social mini-game (some LLM nodes are partially covered, representing that the agent may or may not be in those nodes during an SMG, as the communication could finish before an SMG begins or occur concurrently). See Figure \ref{['smgex']} for a visual example of an SMG.
• Figure 4: Simplified game tree. Each arrow represents a possible combination of control inputs available to the agents. The green box is an example of a subgame starting at $t=1$.
• Figure 5: Trajectories generated by noncommunicative methods in the symmetric doorway. Blue is a grocery agent and red is a hospital agent. Top row is MPC-CBF, middle is SMG-CBF, bottom is GameChat (no LLM). MPC-CBF deadlocks (due to symmetry) and the other methods do not prioritize the hospital agent's urgent task (this happens 50% of the time).

Theorems & Definitions (12)

• Definition III.1
• Definition III.2
• Definition III.3
• Definition III.4
• Theorem IV.1
• proof
• Theorem IV.2
• proof
• Definition IV.1
• Definition IV.2