Robot manipulation improves using symmetry aware vision language models
SAVLA: Symmetry-Aware Vision-Language-Action Models for Robotic Manipulation
Robotics
Summary
Robots that follow language instructions often struggle when objects or scenes look different from their training examples. The authors introduce SAVLA, a new method that helps robots understand and handle objects better by recognizing symmetrical patterns and transforming images into a standard view. This approach allows robots to learn more efficiently from fewer examples and perform more reliably even when scenes are rotated or viewed from different angles. Their tests show SAVLA succeeds more often than previous methods, especially when objects are rotated.
What this means in practice
- •For robotics engineers: Build robotic systems that manipulate objects reliably despite changes in object orientation or camera viewpoints using SAVLA’s symmetry-aware approach.
- •For manufacturing automation teams: Improve robot arm tasks on production lines by enabling action policies that generalize over rotated and angled views of parts without retraining.
Authors
Junle Li, Weixian Waylon Li, Fuxiang Wu, Fusheng Hao, Fengxiang He
Abstract
Vision-language-action (VLA) models have become the dominant paradigm for language-conditioned robot manipulation. However, although images and language instructions inherently encode geometric information, VLAs acquire their spatial competence purely from demonstrations. As a result, they are reliable only within the range of scene poses that the demonstrations cover. We propose SAVLA, an end-to-end symmetry-aware VLA model for robust and data-efficient policy learning. Our approach keeps the pretrained vision-language backbone entirely frozen while combining it with an equivariant flow-matching action head and a learned canonicalizer. The head decomposes its state, action, and conditioning inputs into invariant and equivariant channels, and preserves this typing throughout all of its layers. The canonicalizer transforms oblique-view images into a canonical frame and rotates the geometric conditions consistently. We evaluate our model on LIBERO. Compared with the GR00T N1.5 baseline, SAVLA improves the success rate averaged over all four LIBERO suites by 5.1 points and increases the mean success rate under rotation on LIBERO-Goal from 41.5% to 90.4%.