Negentropy-ai / Ngine

What is Ngine?

Ngine is a Python-based, environment-agnostic engine for embodied AI engineering, unifying diverse robot embodiments like Unitree humanoids and Franka arms across simulators such as Isaac Lab. It delivers a scalable pipeline for benchmark tasks, VR teleoperation data collection, and distributed RL training, streamlining sim-to-real deployment for manipulation and locomotion. Developers get quick setup via shell scripts for teleop, training, and evaluation, with YAML configs handling scenes, robots, and tasks.

Why is it gaining traction?

Built on Isaac Lab-Arena, it stands out by featuring LIBERO benchmarks and Robocasa kitchens out-of-the-box, plus pluggable asset loaders for local/cloud sources across environments. VR teleop with OpenXR and unified MDPs cut data collection time, while distributed setups scale RL across GPUs—ideal for reproducible embodied AI experiments without simulator lock-in.

Who should use this?

Robot engineers training RL policies on humanoids or arms for kitchen manipulation; AI researchers benchmarking sim-to-real transfer on LIBERO tasks; teams needing VR teleop for diverse embodiments like Unitree G1 or LeRobot grippers.

Verdict

Try Ngine if you're in Isaac Lab workflows—its modular pipeline accelerates prototyping despite 60 stars and 1.0% credibility signaling early maturity. Docs are README-focused with examples, but expect tweaks as healthcare benchmarks arrive.

(198 words)