Complex Instruction Following with Diverse Style Policies in Football Games

2025/11/25 by Sun, Chenglu, Shen, Shuo, Hu, Haonan +2
Computer Science · #Artificial Intelligence in Games #FOS: Computer and information sciences #Machine Learning (cs.LG) #Multiagent Systems (cs.MA) #Reinforcement Learning in Robotics #Topic Modeling

paper · doi:10.48550/arxiv.2511.19885

openalex publication_date 2025/11/25 · openalex created_date 2025/11/28 · openalex updated_date 2026/07/28

Abstract

Despite advancements in language-controlled reinforcement learning (LC-RL) for basic domains and straightforward commands (e.g., object manipulation and navigation), effectively extending LC-RL to comprehend and execute high-level or abstract instructions in complex, multi-agent environments, such as football games, remains a significant challenge. To address this gap, we introduce Language-Controlled Diverse Style Policies (LCDSP), a novel LC-RL paradigm specifically designed for complex scenarios. LCDSP comprises two key components: a Diverse Style Training (DST) method and a Style Interpreter (SI). The DST method efficiently trains a single policy capable of exhibiting a wide range of diverse behaviors by modulating agent actions through style parameters (SP). The SI is designed to accurately and rapidly translate high-level language instructions into these corresponding SP. Through extensive experiments in a complex 5v5 football environment, we demonstrate that LCDSP effectively comprehends abstract tactical instructions and accurately executes the desired diverse behavioral styles, showcasing its potential for complex, real-world applications.

Citations

• Noise-Resilient Symbolic Regression with Dynamic Gating Reinforcement Learning
• The Llama 3 Herd of Models
• Contrastive Modules with Temporal Attention for Multi-Task Reinforcement Learning
• Large Language Models as Generalizable Policies for Embodied Tasks
• Language to Rewards for Robotic Skill Synthesis
• PaLM-E: An Embodied Multimodal Language Model
• Do As I Can, Not As I Say: Grounding Language in Robotic Affordances
• Language Models as Zero-Shot Planners: Extracting Actionable Knowledge for Embodied Agents
• Conflict-Averse Gradient Descent for Multi-task Learning
• Human Instruction-Following with Deep Reinforcement Learning via Transfer-Learning from Text
• Multi-Task Reinforcement Learning with Soft Modularization
• Mastering Complex Control in MOBA Games with Deep Reinforcement Learning
• Dota 2 with Large Scale Deep Reinforcement Learning
• Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning
• Language as an Abstraction for Hierarchical Deep Reinforcement Learning
• A Survey of Reinforcement Learning Informed by Natural Language
• From Language to Goals: Inverse Reinforcement Learning for Vision-Based Instruction Following
• Meta-Learning for Multi-objective Reinforcement Learning
• Learning to Understand Goal Specifications by Modelling Reward
• Proximal Policy Optimization Algorithms
• Multi-Objective Deep Reinforcement Learning
• Multi-objective Reinforcement Learning with Continuous Pareto Frontier Approximation Supplementary Material

Related

• Diversity is Strength: Mastering Football Full Game with Interactive Reinforcement Learning of Multiple AIs
• Learning Controllable and Diverse Player Behaviors in Multi-Agent Environments
• Language-Driven Coordination and Learning in Multi-Agent Simulation Environments
• Language Agents with Reinforcement Learning for Strategic Play in the Werewolf Game
• Keep CALM and Explore: Language Models for Action Generation in Text-based Games