EquiVLA: A General Framework for Rotationally Equivariant Vision-Language-Action Models

2026-06-18

AI Summary

This work addresses the limited generalization of existing vision-language-action (VLA) models under varying rotational poses due to a lack of geometric inductive bias. The authors propose the first end-to-end SO(2)-equivariant VLA framework, which constructs a fully rotation-equivariant pipeline by freezing a ViT backbone for visual feature extraction, integrating an approximately equivariant EquiPerceiver module, and employing a strictly equivariant EquiActor action head. Evaluated on the LIBERO benchmark, the method achieves an average success rate of 92.6%, substantially outperforming the baseline of 78.1%. On CALVIN tasks, it attains an average episode length of 4.03 compared to the baseline’s 3.45. Furthermore, real-world robotic experiments demonstrate a significant improvement in task success rate, rising from 54% to 72%.

Abstract

Vision-Language-Action (VLA) models have emerged as a powerful paradigm for generalist robot manipulation, yet they lack geometric inductive biases: policies trained at specific orientations require substantially more data to generalize across rotational configurations. We present \textsc{EquiVLA}, the first general framework for end-to-end $\mathrm{SO}(2)$-equivariant VLA models, applicable to any architecture coupling a frozen vision-language backbone with a flow-matching Diffusion Transformer action head. \textsc{EquiVLA} introduces \textsc{EquiPerceptor}, which produces approximately $\mathrm{SO}(2)$-equivariant visual representations from frozen ViT features; and \textsc{EquiActor}, an exactly $\mathrm{SO}(2)$-equivariant flow-matching Diffusion Transformer action head. Together, they establish an approximate $\mathrm{SO}(2)$ equivariance chain from camera observations to predicted action sequences. Instantiated on GR00T~N1.5 and evaluated across four LIBERO suites, CALVIN ABCD$\to$D, and five real-robot tasks on Mobile ALOHA, \textsc{EquiVLA} achieves $92.6\%$ average success on LIBERO (vs. $78.1\%$ baseline), an average sequence length of $4.03$ on CALVIN (vs. $3.45$), and improves real-robot success from $54\%$ to $72\%$.

Problem

• Vision-Language-Action models
• rotational equivariance
• robot manipulation
• geometric inductive biases
• generalization

Innovation

• rotationally equivariant
• Vision-Language-Action models
• Diffusion Transformer
• flow matching
• geometric inductive bias