Partially Observable Mean Field Multi-Agent Reinforcement Learning with Graph-Attention
The author proposes a novel approach, GAMFQ, using Graph-Attention to select important neighborhood agents for effective multi-agent reinforcement learning.
Decentralized Multi-Agent Reinforcement Learning for Space Robot Motion Planning
The author proposes a decentralized multi-agent reinforcement learning paradigm for trajectory planning and base reorientation tasks for multi-arm space robots. By hierarchically assigning control tasks to different agents, the approach improves exploration efficiency and robustness.
Independent Reinforcement Learning for Cooperative-Competitive Agents: A Mean-Field Perspective
Developing a novel algorithm, MRPG, for achieving Nash equilibrium in General-Sum LQ Mean-Field Type Games.
Multi-Agent Transformer-Accelerated RL for Satisfaction of STL Specifications
Efficiently solving temporally dependent multi-agent problems using transformers.
MA4DIV: Multi-Agent Reinforcement Learning for Search Result Diversification
Introducing MA4DIV, a novel approach using Multi-Agent reinforcement learning for search result diversification, addressing limitations of existing methods.
Bottom-Up Network (BUN): A Novel Approach for Sparse and Decentralized Multi-Agent Reinforcement Learning
The Bottom-Up Network (BUN) approach tackles scalability challenges in multi-agent reinforcement learning by initializing a sparse network that promotes independent agent learning and dynamically establishes connections based on gradient information, enabling efficient coordination while minimizing communication costs.
Exploration-Enhanced Equivariant Graph Neural Networks for Improved Sample Efficiency and Generalization in Multi-Agent Reinforcement Learning
This paper introduces E2GN2, a novel approach leveraging equivariant graph neural networks to significantly enhance sample efficiency and generalization in multi-agent reinforcement learning tasks.
Safe Multi-Agent Reinforcement Learning for Cooperative Robot Navigation Using a Single Target
This paper introduces a novel approach to safe, behavior-based cooperative robot navigation using multi-agent reinforcement learning (MARL) with a centralized training and decentralized execution scheme, requiring only a single target for the formation's centroid and employing model predictive control (MPC) for collision avoidance.
Learning Nash Equilibria in Adversarial Team Markov Games Using Reinforcement Learning
This paper introduces a novel learning algorithm, Independent Stochastic Policy-Nested-Gradient (ISPNG), that enables agents to efficiently learn approximate Nash equilibria in adversarial team Markov games (ATMGs) using only individual rewards and state observations as feedback.
Learning to Balance Altruism and Self-interest Based on Empathy in Mixed-Motive Games: A Novel Multi-Agent Reinforcement Learning Algorithm
This paper introduces LASE, a novel decentralized multi-agent reinforcement learning (MARL) algorithm that leverages a computational model of empathy to balance altruism and self-interest in mixed-motive games, promoting cooperation while mitigating exploitation.
© 2024 by Linnk AI