embryo_lab_locomotion
Collection
6 items • Updated
A locomotion policy for g1 (rough) trained with embryo_lab on Isaac Lab task Isaac-Velocity-Rough-G1-v0 and exported to TorchScript.
Pull this checkpoint straight from the Hub -- no manual download:
from embryo_lab import World
from embryo_lab.robots import G1
from embryo_lab.skills import Locomotion
robot = G1(locomotion=Locomotion("hub:g1/rough"))
with World(robot=robot) as world:
robot.locomotion.move(vx=1.0)
while world.is_running:
world.step()
Locomotion reads the metadata.json deploy contract beside the policy and World auto-pins the physics rate to training, so it drives correctly with no tuning.
metadata.json)
None decimation: 4 action_scale: 0.5['base_lin_vel', 'base_ang_vel', 'projected_gravity', 'velocity_commands', 'joint_pos', 'joint_vel', 'actions', 'height_scan']