Published on October 21, 2026
Oluwaseun Doherty seun-doherty Follow
nvidia
Oyindamola Omotuyi OOyindamola Follow
nvidia
Asier Arranz asiernvidia Follow
nvidia
-
3. Accelerate locomotion training with parallel Isaac Lab environments
-
4. Evaluate the Microduck walking policy and export it to ONNX
-
5. Share policies on Hugging Face and evaluate them in Microduck Arena
-
7. Extend Microduck training to rough terrain and perception tasks with Isaac Lab
Co-authored with Antoine Pirrone, Augustin Crampette, Matthieu Lapeyre, Coralie Deplanne, Anne-Charlotte Passanisi, and Thomas Wolf.
TL;DR
Microduck is an affordable, open-source bipedal robot developed by Pollen Robotics in collaboration with Hugging Face. You can now train Microduck to walk using Isaac Lab and Newton. Test your policies, export them to ONNX, and deploy them directly onto physical hardware. Share your policies on the Hugging Face Hub, benchmark them in the Microduck Arena, and scale up to more advanced tasks with NVIDIA Isaac Lab.
Introduction
Walking policies trained in simulation must be robust enough to handle actuator friction, communication latency, and mechanical backlash on physical robots. NVIDIA Isaac Lab is an open-source framework designed to train policies across thousands of GPU-parallelized simulation environments. Powered by Newton, actuator modeling, and domain randomization, it prepares policies for real hardware. Microduck brings this workflow to…