Simulation testedlocomotionv1.0.0Apache-2.0

Rough Terrain Adaptive Gait

Robust bipedal walking policy trained with severe terrain height variation, uneven steps, and external lateral pushes.

About this behavior

Extends the baseline velocity policy with continuous heightfield domain randomization in MuJoCo Warp. Features higher foot clearance during the swing phase and aggressive compliance response when the leading foot strikes premature obstacles or depressions.

Runtime contract

61-dimension input · 14 joint targets · 50 Hz

Show details

50 Hz / 4×slot: walk

Observation vector61 dimensions · normalizer baked in

Proprioception48D

Positions, velocities, previous actions, gravity, and gyro.

Twist command3D

Forward, lateral, and yaw velocity intent.

Head pose4D

Neck and head orientation targets.

Body pose6D

Six-axis torso offset command.

Action targetsscale: 1×

Left leg5 joints

hip yaw · hip roll · hip pitch · knee · ankle

Neck & head4 joints

neck pitch · head pitch · yaw · roll

Right leg5 joints

hip yaw · hip roll · hip pitch · knee · ankle

Actuator model: Dynamixel XL330 (BAM M6 voltage control law). The beak motor is separate from this fixed locomotion action space.

Verification

Status: Simulation tested
Evidence: Verified in mjlab simulation traversing procedural heightfield terrain and surviving continuous lateral disturbance forces.
Target hardware: MicroDuck v1 (Rockchip RK3566, 14 Dynamixel XL330-M077)
Simulation: mjlab (MuJoCo Warp) at 50 Hz
Note: Higher energy consumption due to elevated swing trajectory.

Compatibility

Robot model: microduck-standard
MJCF model: robot_walk.xml
Terrain: roughflat
Required accessories: None — standard duck setup

Run it on your duck

Use the canonical model and this policy slot to try the behavior locally.

slot: walk

Try the simulation first. Hardware testing assumes a matching Microduck, clear space, and a safe surface. uDuck does not guarantee that a community policy is safe to run.

Deployment stepsShow instructions

1 Download the ONNX model rough_walking.onnx

curl -L "https://raw.githubusercontent.com/pollen-robotics/microduck_rl/develop/docs/models/rough_walking.onnx" -o "/opt/robot/policies/rough_walking.onnx"

2 Add the policy to robotd

[policy]
walk = "/opt/robot/policies/rough_walking.onnx"

3 Restart the control loop

sudo systemctl restart robotd
robotctl health

⌁ Try it in simulation

uv run play Mjlab-Velocity-Rough-MicroDuck

Sources

Upstream repositorypollen-robotics/microduck_rl Training codeMjlab-Velocity-Rough-MicroDuck