An in-depth, 10-chapter tutorial on developing reinforcement learning environments for legged robots in NVIDIA Isaac Lab. This tutorial guides users from running a baseline example with a quadruped (Unitree Go2) to tackling the advanced challenge of humanoid locomotion (Unitree H1).
Each chapter has its own dedicated branch, allowing you to easily follow along or check out the completed code for any part of the tutorial.
➡️ Click here to read the full tutorial on Notion
| Chapter | Title | GitHub Branch |
|---|---|---|
| 1 | Environment Familiarization and Baseline Execution | (Uses built-in code) |
| 2 & 3 | Project Scaffolding & Environment Anatomy | chapter2_3 |
| 4 | Asset Integration and Kinematic Analysis | chapter4 |
| 5 | Implementation and Validation of Kinematics Solvers | chapter5 |
| 6 | Custom Action Integration and Stability Tuning | chapter6 |
| 7 | Advanced Reward Shaping and Configuration Workflows | chapter7 |
| 8 | Adaptive Velocity Curriculum | chapter8 |
| 9 | Bridging the Sim-to-Real Gap with ActuatorNet | chapter9 |
| 10 | From Evaluation to Humanoid Frontiers | chapter10 |
This tutorial was created by Jihoon Moon from the Lab of AI and Robotics at Sungkyunkwan University (SKKU).
Contact: moonjihoon89@gmail.com
This project is licensed under the Apache License 2.0. Please see the LICENSE file for more details.
A significant portion of this tutorial's code is based on or adapted from the official NVIDIA Isaac Lab source code, which is released under the BSD-3-Clause license. This work would not be possible without the foundational efforts of the Isaac Lab development team.