aiqbench analysis
PearlVLA addresses the latency-deliberation trade-off in Vision-Language-Action models by iteratively refining latent action plans using a frozen world model and process-reward reinforcement learning. This approach allows robotic agents to generate low-latency control while incorporating explicit planning benefits without the full computational cost of traditional reasoning methods. The framework demonstrates competitive performance on standard robotics benchmarks, offering a promising direction for efficient embodied AI systems.
PearlVLA: Progressive Embodied Action-Plan Refinement in Latent Space
arXiv:2606.17924v2 Announce Type: replace-cross Abstract: Current Vision-Language-Action (VLA) models face a trade-off between efficient action generation and explicit deliberation. Directly decoding actions from vision-language backbone representations enables low-latency control, whereas textual reasoning, pixel-level subgoals, or world-model evaluation of decoded actions can improve planning but incur substantial latency and computational cost. We propose PearlVLA, a VLA framework that progressively refines a VLM-derived latent plan using feedback from the predicted consequence of each intermediate plan. PearlVLA uses a frozen latent world model (LaWM) pretrained on action-free video. At each refinement round, the current plan produces a continuous latent action code, and the LaWM predicts the corresponding latent visual subgoal. A future-guided plan refiner uses this subgoal to update the plan, so each revision reshapes the next LaWM query. After K rounds, the refined plan is passed once to the host policy's action interface to produce an action chunk. We further introduce Causal Refinement-Grouped Process-Reward RL to optimize latent refinement by comparing rewards from the longer-horizon imagined futures of plan edits made at the same refinement state. Experiments on the LIBERO and RoboCasa benchmarks show that PearlVLA performs competitively against strong existing methods.
Abstract:Current Vision-Language-Action (VLA) models face a trade-off between efficient action generation and explicit deliberation. Directly decoding actions from vision-language backbone representations enables low-latency control, whereas textual reasoning, pixel-level subgoals, or world-model evaluation of decoded actions can improve planning but incur substantial latency and computational cost. We propose PearlVLA, a VLA framework that progressively refines a VLM-derived latent plan using feedback from the predicted consequence of each intermediate plan. PearlVLA uses a frozen latent world model (LaWM) pretrained on action-free video. At each refinement round, the current plan produces a continuous latent action code, and the LaWM predicts the corresponding latent visual subgoal. A future-guided plan refiner uses this subgoal to update the plan, so each revision reshapes the next LaWM query. After K rounds, the refined plan is passed once to the host policy's action interface to produce an action chunk. We further introduce Causal Refinement-Grouped Process-Reward RL to optimize latent refinement by comparing rewards from the longer-horizon imagined futures of plan edits made at the same refinement state. Experiments on the LIBERO and RoboCasa benchmarks show that PearlVLA performs competitively against strong existing methods.
21 pages, 3 figures. Preprint
Robotics (cs.RO); Artificial Intelligence (cs.AI)
arXiv:2606.17924 [cs.RO]
(or arXiv:2606.17924v2 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2606.17924 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 284 words
Related coverage
PearlVLA: Progressive Embodied Action-Plan Refinement in Latent Space
arXiv · Jun 17, 2026
Embodied Robot Manipulation in the Era of Foundation Models: Planning and Learning Perspectives
arXiv · Aug 18, 2026
Import AI 474: Platonic mindspace; TPUs in space; Zhipu starts an outer RSI loop
Import AI · Sep 28, 2026
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026