Preprint
Oct 2026
MixVLA: Adaptive Mixing of Non-Invariant Information for Generalizable Vision-Language-Action Models
Extensive experiments across challenging manipulation settings, including LIBERO, LIBERO-Plus, the RoboTwin perturbation suite, and real-world tasks, demonstrate that MixVLA improves overall zero-shot robustness while retaining strong in-domain performance.
Ping-Rui Zhang, Yu Zhang, Peng-Yuan Wu et al.
· 0 citations