Search papers, labs, and topics across Lattice.
This paper introduces a Feasible Action Neighborhood (FAN)-guided regularizer for vision-language-action (VLA) finetuning, which leverages the property that robotic manipulation states often admit a neighborhood of near-equivalent actions. The FAN regularizer shapes the model's output distribution using a Gaussian prior to promote locally smooth and unimodal predictions around the preferred action. Experiments on reinforced finetuning (RFT) and supervised finetuning (SFT) demonstrate significant improvements in sample efficiency and success rate in both in-distribution and out-of-distribution scenarios.
Exploiting the inherent action tolerance in robotic manipulation lets VLA models learn faster and generalize better.
In real-world robotic manipulation, states typically admit a neighborhood of near-equivalent actions. That is, for each state, there exist a feasible action neighborhood (FAN) rather than a single correct action, within which motions yield indistinguishable progress. However, prevalent VLA training methodologies are directly inherited from linguistic settings and do not exploit the FAN property, thus leading to poor generalization and low sample efficiency. To address this limitation, we introduce a FAN-guided regularizer that shapes the model's output distribution to align with the geometry of FAN. Concretely, we introduce a Gaussian prior that promotes locally smooth and unimodal predictions around the preferred direction and magnitude. In extensive experiments across both reinforced finetuning (RFT) and supervised finetuning (SFT), our method achieves significant improvement in sample efficiency, and success rate in both in-distribution and out-of-distribution (OOD) scenarios. By aligning with the intrinsic action tolerance of physical manipulation, FAN-guided regularization provides a principled and practical method for sample-efficient, and generalizable VLA adaptation.
