Shape Your Body：用价值梯度做多形态机器人设计

Shape Your Body: Value Gradients for Multi-Embodiment Robot Design

AI 导读

研究提出把通用多形态价值函数变成可复用的机器人设计模型：先在大量机器人设计上训练形态感知的策略与价值函数，再用冻结的价值函数作为可微代理，通过价值梯度优化候选形态。单模型最多在 50 个机器人、超 1100 个连续形态参数的设计空间上训练，并覆盖从扰动单机器人到跨形态类别留出机器人的设置。价值梯度还能定位限制性能的设计与控制参数，用于新机器人设计的优化与分析。

正文

View PDF HTML (experimental)

Abstract:We propose to turn generalist multi-embodiment value functions into reusable models for robot design. Instead of running a new reinforcement learning co-design loop for each robot, we first train an embodiment-aware policy and value function across many robot designs. After training, the frozen value function is used as a differentiable surrogate to optimize candidate embodiments through value gradients. We evaluate our approach across different robot design settings, from perturbed single robots to held-out robots across morphology classes, with single models trained on up to 50 robots and design spaces of over 1100 continuous embodiment parameters. Beyond optimizing complete embodiments, we show that value gradients can identify performance-limiting design and control parameters, enabling both the optimization and the analysis of new robot designs.

Subjects:	Robotics (cs.RO); Artificial Intelligence (cs.AI)
Cite as:	arXiv:2606.00702 [cs.RO]
	(or arXiv:2606.00702v2 [cs.RO] for this version)
	https://doi.org/10.48550/arXiv.2606.00702 arXiv-issued DOI via DataCite

Submission history

From: Nico Bohlinger [view email]
[v1] Sat, 30 May 2026 12:21:09 UTC (23,876 KB)
[v2] Tue, 13 Oct 2026 04:14:22 UTC (29,542 KB)

来源：arXiv：cs.AI · arxiv.org