It Takes Two to Tangram: Multi-Agent Object Centric Reinforcement Learning for Assembly
It Takes Two to Tangram: Multi-Agent Object Centric Reinforcement Learning for Assembly
This paper introduces an object-centric reinforcement learning framework where each assembly part is treated as an agent within a graph neural network. The system learns shared policies for continuous control, predicting translational and angular velocities to avoid collisions and deadlocks during planar Tangram assembly. Using procedural generation for diverse configurations, the authors evaluate relational structure and multi-agent optimization in simulation before transferring the policy to sequential execution on a real robot, decoupling high-level coordination from low-level physical control.
The authors present a multi-agent reinforcement learning framework for geometric assembly tasks, demonstrating simulation-to-real transfer on planar Tangram configurations; claims of concurrent assembly and policy decoupling are supported by the provided abstract but specific metrics such as success rates or latency are not reported in the supplied text.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Environment
- SimulationOpen source S4
- code: not_reported
- weights: not_reported
- dataset: not_reported
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ace. 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
Open source S4
object-centric formulation decouples learned assembly coordination from low-level physical control. Project page: https://tangrambench.org/
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org