ACOT-VLA-WM is an advanced Vision-Language-Action robotics framework that uses a predictive world model to generate visual subgoals, improving long-horizon robotic manipulation accuracy and robustness.

Open-source implementation of "RL Token: Bootstrapping Online RL with Vision-Language-Action Models"

Real-time inference engine for openpi

ROS2-native runtime, benchmark, and adapter hub for Vision-Language-Action models.

Unofficial OpenPI extension experiment to build a more complete OpenPI-style VLA engineering stack: pi0.5 semantics, RTC, pi0.6 RECAP/MEM, and pi0.7-inspired world-model ideas.

Try OpenPI closed loop simulation / scenes generation / 3DGS model scenes in genie_sim

Per-Group Error, Not Total MSE. Fine-tuning VLAs on the Toyota HSR. ICRA 2026 Workshop From Data to Decisions, 1st place out of 36.

Physical Intelligence — foundation models for robotics

Deploy Vision-Language-Action models in ROS2 systems with this native runtime, benchmark, and adapter hub.

基于 pi0/OpenPI 的 RM 双臂机器人 MuJoCo 仿真与策略接入项目（由 2025 年 6 月小项目整理，可能存在 bug）。