analiza aiqbench
ViVa addresses a critical gap in robot reinforcement learning by repurposing pretrained video generators to estimate value functions, effectively overcoming the limitations of static vision-language models in capturing temporal dynamics. This approach enables more reliable value estimation for long-horizon tasks by coupling progress assessment with anticipated embodiment dynamics, achieving an 80% average success rate in integrated systems. The work demonstrates that leveraging spatiotemporal priors from video generation offers a promising pathway for improving policy improvement in real-world robotic deployment.
ViVa: A Video-Generative Value Model for Robot Reinforcement Learning
arXiv:2604.08168v2 Announce Type: replace Abstract: Vision-language-action (VLA) models have advanced robot manipulation through large-scale pretraining, but real-world deployment remains challenging due to partial observability and delayed feedback. Reinforcement learning addresses this via value functions, which assess task progress and guide policy improvement. However, existing value models built on vision-language models (VLMs) struggle to capture temporal dynamics and physical interactions, undermining reliable value estimation in long-horizon tasks. In this paper, we propose ViVa, a video-generative value model that repurposes a pretrained video generator to jointly predict future proprioception and a scalar value. By grounding value estimation in anticipated embodiment dynamics, ViVa leverages spatiotemporal priors to intrinsically couple value with foresight beyond static snapshots. ViVa achieves state-of-the-art results in metric-based evaluation across three tasks, producing reliable value signals that accurately track task progress and detect execution errors. Integrated into RECAP, it achieves an average success rate of 80%, highlighting the promise of video-generative models for value estimation.
Abstract:Vision-language-action (VLA) models have advanced robot manipulation through large-scale pretraining, but real-world deployment remains challenging due to partial observability and delayed feedback. Reinforcement learning addresses this via value functions, which assess task progress and guide policy improvement. However, existing value models built on vision-language models (VLMs) struggle to capture temporal dynamics and physical interactions, undermining reliable value estimation in long-horizon tasks. In this paper, we propose ViVa, a video-generative value model that repurposes a pretrained video generator to jointly predict future proprioception and a scalar value. By grounding value estimation in anticipated embodiment dynamics, ViVa leverages spatiotemporal priors to intrinsically couple value with foresight beyond static snapshots. ViVa achieves state-of-the-art results in metric-based evaluation across three tasks, producing reliable value signals that accurately track task progress and detect execution errors. Integrated into RECAP, it achieves an average success rate of 80%, highlighting the promise of video-generative models for value estimation.
Robotics (cs.RO); Artificial Intelligence (cs.AI)
arXiv:2604.08168 [cs.RO]
(or arXiv:2604.08168v2 [cs.RO] for this version)
Text integral sub licența deschisă a publicației · 163 cuvinte