Understanding LLM-Based Bargaining of Driving Agents in Mixed-Autonomy Traffic

Lingxiao Zhou; Di Chen; Jia Li

physics.soc-phOct 30, 2026

Understanding LLM-Based Bargaining of Driving Agents in Mixed-Autonomy Traffic

Authors: Lingxiao Zhou, Di Chen, Jia Li

Organizations: Washington State University · University of California, Davis

Abstract

Large Language Models (LLMs) offer a flexible computational approach for modeling social interactions among heterogeneous agents. This paper develops a novel framework towards characterizing and understanding LLM-based population-level bargaining of shared road space between autonomous vehicle (AV) and human-driven vehicle (HV) agents in mixed-autonomy traffic. Through controlled experiments varying traffic density, agent objectives, and traffic-domain knowledge, we first characterize bargaining outcomes and observe heterogeneous agreement rates across different experimental conditions. To understand how the bargaining process shapes the outcomes, we introduce a behavioral taxonomy that jointly characterizes numerical bidding actions and communicative attitudes. We examine two complementary aspects of social interaction: inter-agent bargaining dynamics and private-to-public behavioral transformation. Results reveal distinct interaction patterns associated with agreement and breakdown, including greater constructive reciprocity in agreement trials and more frequent conflict escalation in breakdown trials. Furthermore, private and public communicative attitudes often differ even when numerical bids remain unchanged. These findings illustrate how LLM-based bargaining simulations can support a structured analysis of social interaction behaviors in mixed-autonomy traffic.

Figures & tables

Figure 1: Overview of the proposed LLM-agent road-share bargaining framework.

Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Condition	AV personality	HV personality
Both-Aggressive	Aggressive	Aggressive
Fair-AV	Fairness-oriented	Aggressive
Fair-HV	Aggressive	Fairness-oriented
Both-Fair	Fairness-oriented	Fairness-oriented

Appendix

Table 1: Agent objective assignments across experimental conditions.

Attitude	Definition
Acknowledge	Recognizes or validates the opponent’s point, concern, or need.
Evidence	Supports a position using facts, domain knowledge, or reasoning without confronting the opponent.
Ignore	Does not meaningfully engage with the opponent’s preceding point and instead continues its own position.
Dismiss	Challenges or rejects the validity or reasonableness of the opponent’s position or reasoning.
Push	Directly attempts to make the opponent change its position or concede.
Threat	Conditions an agent-imposed adverse consequence on the opponent’s failure to change its position.

Appendix

Table 2: Taxonomy of bargaining attitudes.

Figure 2: Overview of bargaining agreement outcomes. (a) Agreement rates across four agent-objective configurations under Knowledge ON and Knowledge OFF, aggregated over all feasible traffic states. (b) Agreement rates across HV-denser, equal-density, and AV-denser traffic-composition regimes, aggregated over all objective and knowledge configurations.

Figure 3: Immediate behavioral responses conditioned on negotiation outcomes. Panels (a) and (b) show attitude-to-attitude responses for agreement and breakdown trajectories, respectively. Panels (c) and (d) show attitude-to-numerical-action responses. Percentages are row-normalized. Sparse categories, particularly Ignore and Threat , are retained for completeness but interpreted cautiously.

Figure 4: Prevalence of representative interaction patterns in agreement and breakdown trajectories. Percentages indicate the proportion of trials within each outcome category that exhibit the corresponding pattern at least once. Numbers in parentheses report the number of trials exhibiting the pattern over the total number of trials in that outcome category.

Figure 5: Representative bargaining trajectories leading to agreement or breakdown. Each stacked bar represents one bargaining round, with segment height indicating the public bid and color indicating the public attitude. The stacking order follows the agents’ speaking order.

Figure 6: Private-to-public attitude transformations. Rows represent attitudes expressed in private notes and columns represent attitudes expressed in public messages. Each row is normalized independently, and each cell reports the conditional percentage and raw number of turns.

Figure 7: Numerical realization of major private-to-public attitude transformations. Rows represent selected semantic transformations, while columns indicate whether the public bid is lower than, identical to, or higher than the corresponding private bid. Percentages are normalized within each attitude transformation, and raw counts are shown in parentheses.

Transition	Recurring Pattern	Private-Channel Expression	Public Realization
Dismiss $\rightarrow$ Push	Pressure / firmness signaling	Negative assessment of the opponent combined with stated intentions to signal resolve, maintain firmness, apply pressure, or encourage reconsideration.	Opponent-directed requests or pressure to change position, often without changing the numerical bid.
Push $\rightarrow$ Dismiss	Indirect pressure through criticism	Stated intention to obtain a concession, pressure the opponent, or encourage reconsideration.	Public criticism of the validity or reasonableness of the opponent’s position rather than a direct request for change.
Dismiss $\rightarrow$ Acknowledge	Strategic softening	Negative assessment of the opponent combined with intentions to “show flexibility,” “signal willingness to negotiate,” or demonstrate willingness to compromise.	Recognition of the opponent’s concerns or perspective, with either an unchanged or lower public bid.
Push $\rightarrow$ Evidence	Strategic justification / reframing	Pressure-oriented intentions involving anchoring, leverage, signaling strength, or encouraging concession.	Reason-based justification of the proposed allocation using traffic density, safety, efficiency, or fairness, usually without changing the bid.
Acknowledge $\rightarrow$ Evidence	Constructive reframing	Recognition of the opponent’s needs, constraints, or reasoning when forming the bargaining position.	Shift in emphasis from recognizing the opponent to explaining and justifying the agent’s own proposal.

Appendix

Table 3: Qualitative interpretation of major private-to-public attitude transformations.

Figure 8: Representative multi-round bargaining trajectories. (a) Case 1: agreement under the Both-Fair condition. (b) Case 2: deadlock under the Both-Aggressive condition.

Multi-Agent Negotiation

Multi-agent negotiation studies how autonomous agents bargain, coordinate, and resolve conflicts over resources, contracts, and shared decisions.

See all 59 papers

Sharing the Road: LLM-Based Bargaining in Mixed-Autonomy Traffic
Nov 6, 2026Lingxiao Zhou, Di Chen, Jia LiIntelligent Transportation Systems Multi-Agent Negotiation
Washington State University · University of California, Davis
Self-Driving Negotiator: An interactive, verifiable benchmark for social negotiation and theory of mind under hidden intent
Jun 13, 2026Ashutosh KumarAutonomous Driving Benchmarks Automated Negotiation
Owl Autonomous Imaging, Inc.
Social Behavior Among Autonomous AI: How Large Language Models Interact in Dynamic Networks
Aug 25, 2026Narges Fardnia, Fatemeh Seyedin, Matthias Becker +2Multi-LLM Collaboration Large Language Model-Based Social Simulation
Gisma University of Applied Sciences, BRAIN Research Center, Potsdam, Germany · Leibniz University Hannover, Germany · University of Cambridge, Cambridge, United Kingdom

Nov 6, 2026J/K move · Enter open · S save

Sharing the Road: LLM-Based Bargaining in Mixed-Autonomy Traffic

Lingxiao Zhou, Di Chen, Jia Li

Washington State University · University of California, Davis

Intelligent Transportation Systems Multi-Agent Negotiation Multi-Agent Coordination Multi-Agent Resource Allocation

Open paper AskPDF arXiv

Mixed-autonomy traffic involves heterogeneous driving populations competing for shared road space. Previous studies have shown the existence and emergence of collective rationality using game-theoretic and reinforcement learning approaches, which rely on explicitly specified behavioral assumptions or reward functions for driving agents. However, how bargaining behaviors between driving agents evolve and lead to collective outcomes based on richer agent characteristics and language-based reasoning remains less understood. To fill this gap, we develop a bargaining framework based on large language models (LLMs) in which population-level autonomous vehicle (AV) and human-driven vehicle (HV) agents negotiate road-space allocations under heterogeneous objectives and private information. Through controlled experiments varying traffic composition, agent objectives, and traffic-domain knowledge, we examine bargaining behavior, coordination, and resource utilization. Results show that negotiated allocations respond to relative traffic density without strictly following proportional sharing, while density-balanced states exhibit greater coordination difficulty. Fairness-oriented objectives substantially improve agreement rates but do not necessarily yield equally high road-space utilization. Furthermore, domain knowledge affects bargaining outcomes differently across objective configurations, shaping both coordination and the role-dependent effects of fairness. These findings demonstrate the potential of LLM-based bargaining for modeling and designing interactions among driving agents in mixed-autonomy traffic.

Abstract

Multi-Agent Negotiation

Explore similar work

Sharing the Road: LLM-Based Bargaining in Mixed-Autonomy Traffic

Self-Driving Negotiator: An interactive, verifiable benchmark for social negotiation and theory of mind under hidden intent

Social Behavior Among Autonomous AI: How Large Language Models Interact in Dynamic Networks