arXiv cs.AI· Laurens Engwegen, Max Weltevrede, Caroline Horsch, Daan Brinks, Wendelin B\"ohmer·· 11 小时前AI 评分24
基于共享模块化递归的上下文 MDP 实现通用形态机器人控制
Shared Modular Recurrence in Contextual MDPs for Universal Morphology Control
AI 导读
研究者提出一种基于 Transformer 的共享模块化递归架构，用于上下文 MDP 下的通用形态机器人控制，可在上下文特征仅部分可用时通过模块化交互恢复信息。在大量 MuJoCo 机器人上评测显示，该架构在四种环境中对动力学、运动学和拓扑均未见的机器人实现了零样本泛化性能的显著提升。
正文 · 原文
Abstract:A universal controller for any robot morphology would greatly improve computational and data efficiency. Steps have been made towards such multi-robot control by utilizing contextual information about the properties of individual robots and exploiting their modular structure in the architecture of deep reinforcement learning agents. When the robots have highly dissimilar morphologies, however, this becomes a challenging problem, especially when the agent must generalize to new, unseen robots. In this paper, we consider the fact that contextual features are often only partially available, but that they can be recovered through modular interactions. This can allow for better multi-robot control and generalization to contexts that are not seen during training. To this extent, we implement a transformer-based architecture with shared modular recurrence and evaluate its (generalization) performance on a large set of MuJoCo robots. The results show a substantial improvement in zero-shot generalization performance on robots with unseen dynamics, kinematics, and topologies, in four different environments.
| Subjects: | Artificial Intelligence (cs.AI); Robotics (cs.RO) |
| Cite as: | arXiv:2506.08630 [cs.AI] |
| (or arXiv:2506.08630v4 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2506.08630 arXiv-issued DOI via DataCite |
Submission history
From: Laurens Engwegen [view email]
[v1]
Tue, 10 Jun 2025 09:44:30 UTC (1,693 KB)
[v2]
Sun, 7 Sep 2025 10:03:07 UTC (1,690 KB)
[v3]
Wed, 8 Jul 2026 09:45:39 UTC (2,768 KB)
[v4]
Fri, 30 Oct 2026 17:07:06 UTC (2,985 KB)
来源：arXiv cs.AI · arxiv.org