Skip to content

论文arXiv 2611.0881413 天前

物中心多智能体强化学习框架实现七巧板装配Sim-to-Real迁移 ​

本文由 AI 根据公开内容整理,可能有误,请以原文为准:It Takes Two to Tangram: Multi-Agent Object Centric Reinforcement Learning for Assembly(arXiv 2611.08814)

核心内容 ​

针对紧密间隙下几何装配易碰撞与死锁的难题,研究提出一种物中心多智能体强化学习框架。该方法将每个装配零件建模为独立智能体,通过共享策略实现连续装配控制。实验基于平面七巧板组装场景,利用程序化生成目标构型与随机初始布局支持多样化训练,验证了该框架在仿真中的并发组装性能及向真实机器人的Sim-to-Real迁移能力。

技术架构采用图神经网络表征零件、目标位姿及空间关系,直接输出各组件的平移与角速度。算法层面使用Per-agent PPO结合集中式图Critic,在单一反馈策略内联合学习运动与时序协调。该物中心公式化解耦了高层装配协调与底层物理控制,使仿真训练的协同策略能无缝适配真机顺序执行模式,有效规避了传统采样规划方法的局限性。

关键看点 ​

  • 物中心RL架构:将装配零件视为独立Agent,利用GNN编码空间关系与目标位姿,直接预测平移和角速度实现连续控制
  • 多智能体优化:采用Per-agent PPO算法配合集中式图Critic,在单一反馈策略中同步学习运动规划与时序协调机制
  • Sim-to-Real迁移:通过物中心公式化解耦高层协调与底层控制,成功将仿真并发组装策略迁移至真机顺序执行场景
  • 数据生成策略:基于程序化生成目标构型与随机初始排列,支撑七巧板装配任务在多样化问题下的训练与泛化评估

工程与产业观察 ​

该工作验证了物中心多智能体RL在精密装配中的Sim-to-Real可行性,其核心在于利用GNN结构化表征解耦任务协调与底层动力学。这种范式降低了真机部署对精确物理模型的依赖,为具身智能在柔性制造、复杂拼装等需精细时空协同场景提供了可迁移的算法参考与开源基准。

英文摘要 (Abstract) ​

Geometric assembly requires fine spatiotemporal coordination to place parts without causing collisions or deadlocks under tight clearances. To address this, we present an object-centric reinforcement learning framework that treats each part as an agent and learns a shared policy for continuous assembly control. A graph neural network represents the pieces, their target poses, and their spatial relationships, and predicts translational and angular velocities for each component. Per-agent proximal policy optimization with a centralized graph critic directly learns motion and temporal coordination within a single feedback policy. We study this formulation through planar Tangram assembly, where pieces must slide and rotate into a given configuration while remaining on the assembly surface. Procedural generation of target configurations and randomized initial arrangements support training and evaluation across diverse assembly problems. Our experiments isolate the roles of relational structure and multi-agent optimization, and compare learned control with sampling-based planning. We demonstrate concurrent assembly in simulation and transfer the policy to sequential execution on a real robot, as our object-centric formulation decouples learned assembly coordination from low-level physical control. Project page: https://tangrambench.org/

作者:Paul Knau{\ss}t, Kay Pompetzki, Nico Bohlinger, Niklas Funk, Cristiana de Farias, Rolf Findeisen, Jan Peters

基于本页技术方向关联的整机参数、核心零部件、实测报告与同赛道选型横评。

外部来源

本页事实对应的公开页面,以原文为准。(点击将离开本站)

  1. arXiv 2611.08814:It Takes Two to Tangram: Multi-Agent Object Centric Reinforcement Learning for Assembly ↗访问 2026-11-09