2026/04/20 by Haoxiang Jie, Yan Yaoyuan, Yaoyuan Yan +5 · 1 voice
Computer Science · Psychology · #Action (physics) #Action recognition #Architecture #Bridge (graph theory) #Embodied cognition #Matching (statistics) #Multimodal Machine Learning Applications #Process (computing) #Reinforcement Learning in Robotics #Reinforcement learning #Social Robot Interaction and HRI #cs.RO
paper · pdf · doi:10.48550/arxiv.2604.17706
openalex publication_date 2026/04/20 · arxiv published 2026/04/20 · openalex created_date 2026/04/22 · arxiv updated 2026/04/24 · openalex updated_date 2026/07/28
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 achieves decent overall performance and surpasses mainstream existing methods, effectively overcoming the fundamental limitations of current VLA models.
RL: Online RL Fine-tuning for Flow-based Vision-Language-Action Models