The paper introduces Preference-based Opponent Shaping (PBOS), a method that incorporates a preference parameter into an agent’s loss function to directly consider an opponent’s loss during strategy updates. By jointly learning strategy and preference parameters, PBOS aims to guide agents toward cooperative or competitive behaviors without relying on simple opponent predictions. Experiments on differentiable games demonstrate that PBOS enables agents to achieve better reward distributions across various environments.
By Xinyu Qiao, Yudong Hu, Congying Han, Weiyan Wu, Tiande Guo
The paper explores a runtime strategy-selection framework where a large language model (LLM) guides a pre‑trained reinforcement learning (RL) policy for non‑player characters (NPCs) in a Unity combat game without altering the underlying policy. Five NPC agents sharing a PPO policy were compared in a baseline setup and an LLM‑augmented setup, where a locally hosted Mistral 7B model assigns one of four tactical tags every five seconds based on live game state. Across 600 episodes against three scripted opponents, the LLM‑augmented agents more than doubled their win rate against a Balanced opponent, improved performance against an Evasive opponent, but struggled against an Aggressive opponent due to over‑reliance on encirclement; analysis of 2,430 strategy selections revealed limited zero‑shot differentiation with the model favoring Surround in 83.8% of cases.
By Hrithika Deepu Nair, Kayvan Karim
arXiv:2608. 09389v1 Announce Type: cross Abstract: This note aims to serve as an entry point to the literature on learning in games, a topic with significant theoretical appeal and a wide range of applications -- from machine learning and data science to economics and beyond.
By Panayotis Mertikopoulos
The paper introduces CASTLE, a framework for fully decentralized multi‑agent reinforcement learning that uses two offline‑trained world models to provide in‑context guidance to independent agents. A Local Dynamics World Model captures each agent’s trajectory dynamics, while a Semantic‑Social World Model predicts short‑horizon task and social outcomes for candidate actions based on counterfactual simulator rollouts. During online learning, these frozen models supply prediction logits to an independent PPO policy, enabling agents to compare potential actions prospectively rather than relying solely on scalar rewards. CASTLE outperforms existing baselines on several multi‑particle benchmark tasks, achieving the highest mean final scores across 30 seeds.
By Fernando Martinez, Tao Li, Yingdong Lu, Juntao Chen
arXiv:2608. 09128v1 Announce Type: cross Abstract: LLM agents are increasingly deployed in multi-agent social settings where they must cooperate, negotiate, and adapt to other agents.
By Keyu He, Xuhui Zhou, Maarten Sap
Learn2Play Bench is a new benchmark that tests how well large language model agents learn from experience in unfamiliar, text‑based games with novel or counterintuitive rules. The benchmark provides reproducible feedback, automatic scoring, and varied game instances to evaluate learning across repeated attempts and transfer to new situations. Findings show that retaining full action records aids learning, human players outperform agents, and the choice of harness significantly impacts performance and inference cost.
By Yibo Li, Jinhang Qiu, Zhi Zheng, Qianyun Guo, Jiaying Wu, Shuo Ji, Bryan Hooi
The paper investigates how to infer agents’ underlying incentives in strategic multi‑agent settings, where traditional inverse reinforcement learning (IRL) may fail to produce rewards that guide optimal play. It introduces the concept of strategic identifiability and shows that learning from a fixed pair of policies can be insufficient regardless of sample size. To address this, the authors propose an active IRL algorithm for two‑player repeated Stackelberg games, combining maximum‑likelihood estimation with experimental design to achieve finite‑sample guarantees on sub‑optimality, and validate the approach in an anti‑poaching scenario.
By Antoine Bergerault, Cyrille Kone, Giorgia Ramponi
arXiv:2608. 11658v1 Announce Type: cross Abstract: Many reinforcement learning systems, from fleet management to traffic signal control, must serve an objective that changes dynamically after deployment, and retraining a policy for each new objective is prohibitively expensive.
By Zijian Zhao, Sen Li
arXiv:2503. 13077v2 Announce Type: replace Abstract: Multi-agent reinforcement learning has shown promise in learning cooperative behaviors in team-based environments.
By Amir Baghi, Jens Sj\"olund, Joakim Bergdahl, Linus Gissl\'en, Alessandro Sestini
The paper investigates how large language models (LLMs) can be trained to cooperate in multi‑agent social dilemmas. It shows that standard reinforcement learning often leads LLM agents to defect and exploit others, even when they have cooperative priors. By adapting the Advantage Alignment algorithm and introducing a group‑relative baseline, the authors fine‑tune LLMs to achieve higher collective payoffs and resist exploitation, and they present a new Trust‑and‑Split environment that requires natural language communication for optimal outcomes.
By Dereck Piche, Mohammed Muqeeth, Milad Aghajohari, Juan Duque, Michael Noukhovitch, Aaron Courville
EpiAgent introduces an explicit, learnable epistemic layer for large language model agents, enabling them to maintain and update beliefs about other agents’ private information. The framework splits the policy into Epistemic Update and Action Planning stages, optimized via a bi‑level Epistemic Self‑Play algorithm that rewards accurate belief predictions. Experiments in mixed and competitive settings show EpiAgent outperforms prompting and RL baselines, highlighting the benefits of dedicated epistemic training for belief inference and action planning.
By Xuanfa Jin, Xinyu Cui, Zhijian Ma, Xue Yan, Jian Zhao, Haifeng Zhang, Jun Wang
AUSO (Action-level Unified Skill Optimization) is a method that unifies skill learning and skill use through a progressive, action-aware optimization process. It starts by jointly learning from teacher guidance and environmental outcomes, then shifts to outcome-based policy optimization, and finally evaluates each action under skill-conditioned and skill-free contexts to strengthen beneficial skill-sensitive actions while suppressing harmful ones. Experiments on ALFWorld, WebShop, and SearchQA demonstrate that AUSO consistently improves agent performance and out-of-distribution generalization compared to competitive baselines.
By Huizu Lin, Chengkai Huang, Tianqi Gao, Tao Huang, Daijiao Liu, Tongxin Li, Xiaoyan Sun, Lina Yao