openpi

Open-source implementation of "RL Token: Bootstrapping Online RL with 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.

Real-time inference engine for openpi

Official implementation of PhaseLoRA: control-regime-conditioned low-rank adaptation for continuous-action vision-language-action policies, built on OpenPI.

Deployment-focused fork of OpenPI for LimX TRON2 manipulation — pi0.5 policy serving, task fine-tuning, TRON2 transforms, and real-robot client examples.

[NeurIPS 2026] Efficient VLA inference under multi-task parallelism

π₀ / π₀-FAST / π₀.₅ vision-language-action models adapted to Xense platforms — fine-tuning and real-world deployment on BiARX5, BiFlexiv and XTac-UMI dual-arm rigs.

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

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.

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

End-to-end open-source stack for real Franka robots: realtime control, GELLO/VR data collection, policy training, and guarded deployment.

An end-to-end tour of embodied AI world models: Closed-loop VLA simulation, Real2Edit2Real spatial scene editing, generalization benchmarking, and in-world-model RL self-evolution using GE-Sim 2.0 & OpenPI π₀.₅.

ArcheBase Physical AI 开放电子书：VLA、世界模型、机器人学习、控制、评测与开放数据实践

IMLE-VLA: Fast Single-Step Action Generation for Vision-Language-Action Policies (IROS 2026)

Reproducible AMD ROCm evaluation and full three-camera recording for BiGym, OpenPI JAX, MuJoCo and 3DGS.

Team Dexterity: training pi0.5 (openpi, JAX) for the 2026 BEHAVIOR Challenge -- code, step-by-step guide, model and results

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.

Kuavo Scene2 VLA: rosbag-to-LeRobot data pipeline, OpenPI π0.5 LoRA training, and ROS/MuJoCo closed-loop deployment

Learn to reproduce OpenPI pi05 LIBERO_10 LoRA fine-tuning step by step: data splits, cloud training, WSL deployment, and rollout evidence. | 手把手复现 OpenPI pi05 的 LIBERO_10 LoRA 微调：涵盖数据划分、云端训练、WSL 部署与推理视频及日志。

Fine-tuning π0-FAST on an RTX 4070 using synchronized camera and joint demonstrations from a custom 4-DOF teleoperation arm, with LeRobot conversion, policy serving, open/closed-loop evaluation, and a real-time dashboard.