The paper investigates task collapse—a failure mode where online RL fine‑tuning of a pretrained flow‑matching vision‑language‑action policy erodes performance on individual tasks—using a 450M‑parameter SmolVLA policy on LIBERO‑10. Three exploration‑noise strategies are compared: a fixed noise scale, a learned noise network, and an uncertainty‑gated controller that reallocates exploration based on novelty and competence signals without task labels. The uncertainty‑gated controller prevents task collapse across all tested seeds, whereas the other two approaches consistently cause collapse, demonstrating its effectiveness in preserving task performance during fine‑tuning.
By Mehmet Turan Yard{\i}mc{\i}, Yunus Emre \c{C}o\u{g}urcu
arXiv:2610.25043v2 Announce Type: replace
Abstract: Flow-matching vision-language-action (VLA) models generate an action chunk by iteratively denoising Gaussian noise, but typically execute only a sh...
By Sicheng He, Yi Pan, Qiuting Yu, Jiaming Huang, Qi Lu, Shuyan Wu, Hongyan Feng, Xuanyu Liu, Jun Zhang, Jie Zhang, Wenqi Zhang, Xuhong Zhang, Jianwei Yin
arXiv:2604.05656v2 Announce Type: replace
Abstract: Vision-language-action (VLA) models with flow-matching action heads, such as {\pi}0 and {\pi}0.5, have become a standard recipe for generalist robo...
By Wuyang Luan, Junhui Li, Weiguang Zhao, Rui Ma, Tieru Wu
CF‑VLA introduces a two‑stage coarse‑to‑fine approach for vision‑language‑action policies, replacing multi‑step sampling with a coarse initialization that constructs an action‑aware starting point and a single‑step refinement that corrects residual errors. The coarse stage learns a conditional posterior over endpoint velocity to transform Gaussian noise into a structured initialization, while the fine stage performs a fixed‑time refinement. Experiments on CALVIN and LIBERO demonstrate that CF‑VLA achieves a strong efficiency‑performance trade‑off, reducing action sampling latency by 75.4 % and achieving an 83.0 % real‑robot success rate, outperforming existing NFE=2 methods and matching or surpassing NFE=10 baselines.
By Fan Du, Feng Yan, Jianxiong Wu, Xinrun Xu, Weiye Zhang, Weinong Wang, Yu Guo, Bin Qian, Zhihai He, Fei Wang, Heng Yang
arXiv:2606. 14801v1 Announce Type: cross Abstract: Flow-matching and diffusion policies are expressive action generators, but optimizing them with temporal-difference reinforcement learning (RL) remains difficult.
By Yifan Ruan, Chenyang Cao, Andreas Burger, Ali Pesaranghader, Kaveh Kamali, Jaehong Kim, Nandita Vijaykumar, Alan Aspuru-Guzik, Igor Gilitschenski, Nicholas Rhinehart
arXiv:2611.01215v2 Announce Type: replace-cross
Abstract: Diffusion-based vision-language-action (VLA) models cascade pretrained vision-language models (VLMs) with diffusion action experts, but typic...
By Xicheng Gong, Wenxuan Song, Chengze Liu, Qiwei Li, Qize Yu, Jiaolong Yang, Haoang Li, Yadong Mu
IMLE‑VLA replaces the iterative action head in vision‑language‑action policies with a single‑step conditional generator trained via conditional Implicit Maximum Likelihood Estimation (cIMLE). This eliminates multi‑step sampling, boosting inference frequency by 3.67× (55 Hz vs. 15 Hz) and achieving the highest average success rate (98.0 %) on the 40‑task LIBERO benchmark while maintaining robustness under perturbations. Real‑world tests on a Franka Emika Panda show smoother, faster motions and a 3.9×–6.6× reduction in inference time per episode.
By Kian Hosseinkhani (Simon Fraser University), Qinhe Peng (University of Pennsylvania), George Shramko (Simon Fraser University), Mehran Aghabozorgi (Simon Fraser University), Jianing Qian (University of Pennsylvania), Tristan Engst (Simon Fraser University), Alireza Moazeni (Simon Fraser University), Dinesh Jayaraman (University of Pennsylvania), Ke Li (Simon Fraser University, Canada CIFAR AI Chair)
arXiv:2607.08837v4 Announce Type: replace-cross
Abstract: Exploration is essential to RL since a policy cannot improve by repeatedly sampling the behaviors it already prefers. Standard methods inject...
By Sunshine Jiang, John Marangola, David Zhang, Raghuram Kowdeed, Ruiyang Luo, Nitish Dashora, Richard Li, Pulkit Agrawal, Zhang-Wei Hong
arXiv:2606. 08602v1 Announce Type: cross Abstract: We present an online reinforcement learning (RL) algorithm for fine-tuning flow-matching policies in continuous-control problems.
By Boshu Lei, Kostas Daniilidis, Antonio Loquercio
arXiv:2606. 31846v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models offer a promising framework for robotic manipulation by connecting language instructions, visual observations, and continuous control.
By Lang Cao, Renhong Chen, Luyi Li, Peng Wang, Mofan Peng, Yitong Li
HybridFlow is a generative policy for robotic manipulation that uses a three‑stage inference procedure requiring only two network function evaluations (2‑NFE). The policy first generates a coarse action trajectory with a Global Jump based on MeanFlow, then refines the state using a parameter‑free ReNoise interpolation, and finally performs a Local Refine to query the instantaneous‑velocity limit. Experiments on RoboMimic and five real‑robot settings show that HybridFlow achieves high success rates and improves task performance over a 16‑step Diffusion Policy while reducing action‑generation latency by roughly eightfold.
By Zhenchen Dong, Fulin Chen, Jinna Fu, Jiaming Wu, Qingran Wu, Shengyuan Yu, Hongyu Yu, Yide Liu
arXiv:2607. 08837v1 Announce Type: cross Abstract: Exploration is essential to RL since a policy cannot improve by repeatedly sampling the behaviors it already prefers.
By Sunshine Jiang, John Marangola, David Zhang, Raghuram Kowdeed, Ruiyang Luo, Nitish Dashora, Richard Li, Pulkit Agrawal, Zhang-Wei Hong