Microduck RL Training Library Open Source: Head Tracking and Backlash Simulation Tricks
Thom_Wolf · x · 2026-08-28
Pollen Robotics released the reinforcement learning training library microduckrl for the bipedal robot Microduck. Built on MuJoCo Warp with PPO, the project includes the full sim2real recipe, covering BAM actuator physics, domain randomization, and backlash simulation. The AGENTS.md document shares concrete reward design insights, such as using EMA to smooth head tracking error (to prevent oscillation caused by the head being 38% of the weight) and modeling motor backlash by adding an unactuated hinge in series with the motor.
Related event: Pollen Robotics open-sources Microduck bipedal robot RL training stack(5 posts)→
More from Embodied
- How Microsoft optimized Windows for Nvidia's Arm-based RTX Spark laptops — The Verge AI · 2026-10-15
- John Deere's second-gen autonomous tillage: 16 cross-checking cameras beat sensor drift — lukas_m_ziegler · 2026-10-15
- Europe's Application-Layer Bet: Can the Rotterdam Humanoid Center Deliver? — ingliguori · 2026-10-15
- VC thesis 'Data at the Edge': cheap sensors plus AI are rebuilding the internet's data flywheel offline — rebeccakaden · 2026-10-15
- PingpongDuo shows humanoid robots playing doubles table tennis — arankomatsuzaki · 2026-10-15
- Robotics data gap: just 300K hours of manipulation data globally vs 100M+ needed, $3B spend projected — 0xSammy · 2026-10-15