Precise Manipulation with Efficient Online RL

We extract an RL Token from VLA models to enable fast online RL and improve throughput on precise tasks with few hours of data.

Precise Manipulation with Efficient Online RL Precise Manipulation with Efficient Online RL Published March 19, 2026 Email research@physicalintelligence.company Charles Xu, Jost Tobias Springenberg, Michael Equi, Ali Amin, Adnan Esmail, Sergey Levine, Liyiming Ke Paper RLT .pdf Loading… Our models can follow diverse instructions and perform a wide range of tasks, from folding laundry and making coffee to cooking a grilled cheese sandwich . But for many applications, broad competence is not enough: the hardest parts of physical tasks require precision, dexterity, and speed. Picking up a screwdr

saved by

• Information Flow
• Vincent Huang
• KRUPAL VIRANI
• Anya Singh
• Alex K
• Anand Majmudar

related reading

• [2602.10556] LAP: Language-Action Pre-Training Enables Zero-shot Cross-Embodiment Transferarxiv.org
• A VLA that Learns from Experiencepi.website
• A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulationtoyotaresearchinstitute.github.io
• SimpleVLA-RL: Scaling VLA Training via Reinforcement Learningalphaxiv.org
• Sparse Autoencoders Reveal Interpretable and Steerable Features in VLA Modelsarxiv.org
• [2206.11795] Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videosarxiv.org
• Emergence of Human to Robot Transfer in Vision-Language-Action Modelspi.website
• Real-Time Action Chunking with Large Modelspi.website
• Dexterous Robotic Foundation Modelsyoutube.com
• Emergent Dexterity via Diverse Resets and Large-Scale Reinforcement Learningarxiv.org
• A VLA with Open-World Generalizationpi.website
• [2103.06326] S4RL: Surprisingly Simple Self-Supervision for Offline Reinforcement Learningarxiv.org