JevVLA improves robot task actions using a shared energy model
JevVLA: Jev-Inspired Energy-Based Adaptation for Vision-Language-Action Policies
Robotics
Summary
Robots that follow instructions often choose from preset actions or decide on movements step-by-step. The authors found a way to improve both choices using a single system that scores how good each possible action is. This system learns from feedback or examples without changing the original robot brain. By training on multiple tasks, the new system helps robots perform better in various test scenarios.
What this means in practice
- •For robotics engineers: Adapt robot control policies to improve pick-and-place and motion tasks by integrating feedback with a lightweight energy scoring system.
- •For industrial automation teams: Enhance robotic manipulation accuracy and adaptability in manufacturing settings by applying a shared energy-based action preference model.$Commercial implications: This enables more reliable and flexible robotic assembly and handling systems that can be marketed to manufacturing industries.
Authors
Wei Wang, Wenqiao Zhang, Zheqi Lv, Zheng Dong, Yingjie Niu, Jiaqi Zhu, Jun Xiao, Chao Li, Yueting Zhuang
Abstract
Vision-language-action (VLA) policies can select among discrete candidates for pick-and-place manipulation or generate continuous action chunks for motion control. Improving both forms of decision making through a common mechanism while retaining the capabilities of a pretrained policy is an important adaptation problem. Inspired by Jev's structured, probabilistic decision interface, we propose JevVLA, a lightweight energy-based adaptation framework for VLA policies. JevVLA instantiates this inspiration by modeling the relative quality of action proposals: with the base policy frozen, it learns a differentiable scalar energy conditioned on observations, task context, and actions, using execution feedback, expert corrections, or demonstration-derived preferences. JevVLA uses this single learned action preference function in two ways: its values compare discrete candidates, while its action gradients correct continuous actions or guide their generation. Candidate selection and continuous control thus share one trainable decision core. Each benchmark uses independently trained weights and fixed input/output encodings, while retaining the same core network with approximately 80,000 trainable parameters. On the held-out test sets, JevVLA improves task success over the respective matched base-policy controls by 1.28, 2.31, 1.60, 1.52, and 1.28 percentage points on MetaWorld MT50, VIMA, LanguageTable, LIBERO Plus, and CLIPort, respectively. These results support the feasibility of a lightweight, Jev-inspired energy model for adapting both discrete action selection and continuous action guidance. The JevVLA implementation will be open-sourced at https://github.com/ZJU4EmbodiedAI/JevVLA.