RLinf /RLinf

RLinf: Reinforcement Learning Infrastructure for Embodied and Agentic AI

★ 5,654⑂ 845👁 17⑃ 1,409● 438▤ 4
PythonApache-2.0Created 2025-08-14
Download: RLinf v0.4 ReleaseRLinf-0.4 ↓0.42026-10-20
+3
Daily
5,654
stars
845
forks
17
watchers
1,409
prs
438
issues
4
releases

README

RLinf: Reinforcement Learning Infrastructure for Embodied and Agentic AI


RLinf is a flexible and scalable open-source RL infrastructure designed for Embodied and Agentic AI. The 'inf' in RLinf stands for Infrastructure, highlighting its role as a robust backbone for next-generation training. It also stands for Infinite, symbolizing the system’s support for open-ended learning, continuous generalization, and limitless possibilities in intelligence development.


  



 What's NEW!

- [2026/11] 🎉 The RLinf website is live. Read our blog posts and browse community events.
- [2026/11] 🔥 RLinf supports MuJoCo Warp for batched GPU simulation on five classic locomotion tasks with an MLP + PPO recipe. Doc: MuJoCo Warp.
- [2026/11] 🔥 RLinf supports LeKiwi, a three-wheel omnidirectional mobile base, with per-part bring-up tools and a registered real-world drive environment. Doc: LeKiwi.
- [2026/11] 🔥 RLinf supports dual-arm ARX X5 data collection with Pico teleoperation. Doc: ARX X5.
- [2026/11] 🔥 RLinf supports DiffusionART fine-tuning for SD3 image generation and Wan2.2 video generation with OCR rewards, reaching the same reward threshold in about 30% less wall-clock time and a third fewer trajectories than NFT on Wan2.2. Doc: DiffusionART.
- [2026/11] 🔥 RLinf supports LIBERO-Mem and MEM-inspired visual short memory for OpenPI SFT and evaluation, raising strict LIBERO-Mem success from 20.0% to 35.5% over the current-frame baseline. Docs: LIBERO-Mem, OpenPI Short Memory.
- [2026/11] 🔥 RLinf supports RL on π₀.₅ in GenieSim with PPO and GRPO. Doc: RL with GenieSim.
- [2026/11] 🔥 RLinf supports RoboDojo, an Isaac Sim-based dual-arm manipulation benchmark for ARX-X5, exposed as a Gym environment for VLA policy evaluation: the official π₀.₅ checkpoint can be converted to OpenPI format and evaluated over 55 predefined scene layouts. Doc: RoboDojo.
- [2026/11] 🔥 RLinf supports SO-101 joint-space demonstration collection, OpenPI π₀.₅ supervised fine-tuning, and local evaluation. Doc: SO-101.
- [2026/11] 🔥 RLinf supports Psi0 VLA and the SIMPLE humanoid simulator. Docs: Psi0, SIMPLE.
- [2026/11] 🔥 RLinf supports PhyAI as an in-process rollout backend for π₀.₅ PPO training on LIBERO, including hot weight synchronization, and cutting π₀.₅ prediction time on the rollout critical path from 142.6 s to 68.7 s on four H800 GPUs. Doc: PhyAI rollout backend.
- [2026/10] 🎉 RLinf v0.4 is released, taking RL out of simulation and into agile control of the physical world. Highlights: a modular real-robot system (rlinf.robotics), real-robot SOTA post-training algorithms (RLT, STEAM, OPD, RTC), full-modality models (native PyTorch OpenPI π₀ / π₀.₅, Cosmos3, FastWAM, π₀-FAST, Evo-1, MolmoAct2, SD3 / Wan2.2), and eight accelerator backends. Release notes: RLinf v0.4.
- [2026/10] 🎉 RLinf introducing CEL (Collective Experience Learning): robot fleets improve a shared policy from their own deployment experience without additional expert demonstrations or corrective actions, validat

Top issues

Discussions

Latest release

RLinf v0.4 Release v0.4 2026-10-20
🚀 RLinf v0.4 is here. Reinforcement learning steps out of simulation and into agile control of the physical world.

v0.3 connected the path from simulation to real robots. v0.4 is the step toward deploying embodied RL at real-world scale. The release rebuilds the stack around three pillars: a modular real-robot system, high-efficiency real-robot algorithms, and full-modality models on a heterogeneous compute base.

- 🧩 Modular real-robot system (rlinf.robotics): arms, dexterous hands, cameras, mobile bases, and teleoperation devices become composable blocks. Hardware drivers are fully deco

Snapshot: 2026-11-16T09:23:14Z · Data license Apache-2.0 · JSON