Search papers, labs, and topics across Lattice.
This paper introduces DiMaS, a novel Distribution-Matching Steering strategy designed for flow-matching vision-language-action (VLA) models, addressing the challenge of fine-grained behavioral control in robotic manipulation. Unlike traditional methods that rely on linear steering, DiMaS transports between representation distributions, enabling more effective behavioral control across various tasks. The authors demonstrate that this approach not only enhances control in state-of-the-art VLAs but also provides insights into the limitations of classical linear steering in the context of visuomotor tasks.
Traditional linear steering methods fail in robotic manipulation, but DiMaS reveals a more effective way to control behavior by matching representation distributions.
Flow-matching-based vision-language-action (VLA) models have emerged as powerful policies for robotic manipulation, yet a critical capability remains underexplored: fine-grained behavioral control, the ability to govern how a robot performs a task by intervening on its internal representations. Representation steering is a well-established interpretability tool for language and vision-language models, where behavioral features are typically encoded as linear directions, but we show that these classic methods fall short in VLAs. We propose DiMaS, a Distribution-Matching Steering strategy tailored to flow-matching VLAs, which transports between representation distributions rather than shifting along a fixed direction, and show that it effectively controls behavior across two state-of-the-art VLAs. We further examine the generalizability of this strategy as the tasks it is learned from and evaluated on grow increasingly dissimilar, characterizing where behavioral control transfers and where it weakens. Finally, through an analysis of the representation structure of the action expert, we explain why classical linear steering falls short in the visuomotor setting: behavioral features are linearly decodable but not linearly steerable, which motivates the distribution-matching design of DiMaS. Our code is publicly available at https://github.com/pegah-kh/dimas, with additional results and videos at https://pegah-kh.github.io/dimas/
