Downloads · 30 days
55
34% of all-time downloads
microduck-moonwalk-backward-55e6af
fffiloni/microduck-moonwalk-backward-55e6af
microduck-moonwalk-backward-55e6af is a reinforcement learning model from fffiloni. Use it for the reinforcement learning task on the model card, and read the license before you ship it in a product. It is set up for microduck.
Generated with Microduck Motion Lab from Pollen Robotics' microduckrl sim-to-real stack.
All-time downloads
163
Public
Repo size
5.8 MB
Likes
2
Public
Repo makeup
Click a slice to open those files.
.pt4.8 MB · 83%
At a glance
- Task
- Reinforcement Learning
- Library
- microduck
- Access
- Public
- Created
- Aug 29, 2026
- Updated
- Aug 29, 2026
- SHA
- 951073cb
Model card
From the Hugging Face model README
Moonwalk Backward — Microduck policy
Generated with Microduck Motion Lab from Pollen Robotics' microduck_rl sim-to-real stack.
Prompt
Make Microduck perform a moonwalk-style backward gliding gait. It should move backward smoothly while alternating its feet, create the visual illusion of sliding rather than ordinary backward walking, keep its upper body stable and upright, avoid jumps or large vertical motion, maintain balance throughout, and finish in a stable standing pose after a few seconds.
Reproducibility
- Task:
Mjlab-MotionLabMoonwalkBackward-Flat-MicroDuck - Base recipe:
velocity microduck_rlrevision:d424a0c899f6b33cbd3daeb279913134349c0b63- Parallel environments: 4096
- PPO iterations: 4000
- Compiler:
recipe-plan-adapter-v0.10 - Final repo:
fffiloni/microduck-moonwalk-backward-55e6af
Deployment checks
- Official
scripts/export.py: passed - Shared actor observation contract: 61D
- Policy action output: 14D
- CPU MuJoCo/ONNX rehearsal: passed
- Automatic motion-quality gate: needs review
- Preview recipe: velocity
- Preview driver: velocity
- Semantic match unverified: True
- Maximum joint motion: 14.737863554636114°
- XY displacement: 0.33397904866706285 m
- Ball displacement: 0.0 m
Files
policy.onnx— deployment policy exported by upstreamscripts/export.pypreview.mp4— headless CPU MuJoCo deployment rehearsalthumbnail.webppreview_metrics.jsonmotion_spec.jsonresolved_plan.jsoncompiled_motion.jsontraining_result.jsoncheckpoints/model_final.pt
A simulation preview is not a guarantee of safe real-robot behavior. Validate conservatively before deploying to hardware.
- Task
- Reinforcement Learning
- Library
- microduck
- Created
- Aug 29, 2026
- Updated
- Aug 29, 2026