microduck-songkran-duck-tug4
tug-of-war control policy for the microduck robot
1 min readupdated
This model is a perpetual control policy for the microduck robotics platform (featuring a 61-D observation space, 14 actions, and a 50 Hz frequency) that runs continuously until stopped. Trained for the Mjlab-TugOfWar-MicroDuck task, the policy loads directly into the robot's slot. An observation normalizer is built right into the ONNX file, allowing the model to accept raw data. The development is intended for researchers and engineers working with robotics hardware and simulations.
Key features
- Perpetual control policy for the microduck platform (50 Hz)
- Observation normalizer built into the policy.onnx file
- Simulation-based training using the Mjlab-TugOfWar-MicroDuck approach
- Full training reproducibility via fixed code and seed
What you can do
- Deploy the policy onto a compatible microduck robot using the robotctl utility
- Reproduce model training with specified parameters using the microduck-challenges repository
Example
sudo robotctl policy load DenRakEiw/microduck-songkran-duck-tug4Read more
DetailsTask: Mjlab-TugOfWar-MicroDuck
Repository: microduck-challenges (commit 6c0b5d452)
Checkpoint: 8900, seed: 11
Base model: mjlab-microduck 0.1.0 @ 1efd853be










Comments