pollen-robotics/microduck_rl +1,122 ★ this week + Follow

What it actually is

This repository provides reinforcement-learning training environments for Microduck, a ~800 g, ~25 cm bipedal robot from Pollen Robotics. It is built on mjlab (MuJoCo Warp) and trains policies with PPO at 50 Hz; task configurations such as make_microduck_roller_standup_env_cfg, make_microduck_spin_env_cfg, make_microduck_ground_pick_env_cfg and make_microduck_velocity_env_cfg define a registry of environments (walking, standing up, sitting, roller-skating, ball-kicking, forward rolls). Trained policies are exported to ONNX and deployed on the physical robot via the runtime in the separate pollen-robotics/microduck repo. It is aimed at robotics/RL engineers doing sim2real work on small bipedal hardware.

Why it's growing

The star jump (1,122 in a week on a repo created 2025-12-06) is not explained by release notes — none are fetched/published — so the growth is better read as attention driven by the linked hero video/demo content (walking, standup, roulade, roller skating) and the README's explicit sim2real pitch rather than by a versioned milestone. Recent commit activity is substantial and feature-bearing (actuator physics, robot model updates, a policy-publishing workflow), which is consistent with an actively maintained project, but the commit log alone doesn't establish causation for the star spike.