Multi-Agent Reinforcement Learning for Multi- Robot Intelligent Fixture Planning

Fixture layout planning is critical for securely holding components during production processes. An optimal fixture arrangement minimizes surface deformation and prevents crack propagation, thereby maintaining the structural integrity of components. Traditionally handled by engineers, fixture planning has grown too complex for manual methods alone. Conventional optimization often gets stuck in local optima, limiting effectiveness. While machine learning offers improvements, it demands costly, labeled data. This paper proposes a multi-agent reinforcement learning framework with team decision theory. The approach enables agents to learn collaboratively, improving fixture planning without heavy data reliance by simulating fixture placement on a flexible surface to minimize deformation under uniform pressure. Multiple agents select fixture pairs, with deformation estimated using plate bending theory. The environment supports reinforcement learning and highlights the benefits of strategic, informed placements.