1 paper · 1 filter
Siting Wang, Xiaofeng Wang, Zheng Zhu +7
Flow-based vision-language-action (VLA) models excel in embodied control but suffer from intractable likelihoods during multi-step sampling, hindering online reinforcement learning…