OmniVLA-RL: A Vision-Language-Action Model with Spatial Understanding and Online RL
This work addresses spatial understanding and training stability issues in embodied AI, offering a new architecture and training method that outperforms existing VLA models.
OmniVLA-RL introduces a Mix-of-Transformers architecture and Flow-GSPO training to improve spatial perception and action precision in VLA models, achieving state-of-the-art results on LIBERO and LIBERO-Plus benchmarks.
Visual-Language-Action (VLA) models represent a paradigm shift in embodied AI, yet existing frameworks often struggle with imprecise spatial perception, suboptimal multimodal fusion, and instability in reinforcement learning. To bridge these gaps, we propose OmniVLA-RL, a novel architecture that leverages a Mix-of-Transformers (MoT) design to synergistically integrate reasoning, spatial, and action experts. Furthermore, we introduce Flow-GSPO, which reformulates flow matching as a Stochastic Differential Equation (SDE) process and integrates it with Group Segmented Policy Optimization (GSPO) to enhance action precision and training robustness. Extensive evaluations on the LIBERO and LIBERO-Plus benchmarks demonstrate that OmniVLA-RL significantly outperforms state-of-the-art methods, effectively overcoming the fundamental limitations of current VLA models.