Importing PACE
The intended public imports match upstream PACE:
from pace_sim2real import (
CMAESOptimizer,
PaceCfg,
PaceSim2realEnvCfg,
PaceSim2realSceneCfg,
)
from pace_sim2real.utils import PaceDCMotor, PaceDCMotorCfg, bind_environment
import pace_sim2real.tasks
pace_sim2real.tasks registers Isaac-Pace-Anymal-D-v0 with mjlab's task
registry. Create it using mjlab.tasks.registry.load_env_cfg and
mjlab.envs.ManagerBasedRlEnv; no Isaac Sim application launcher is needed.
PACE never monkey-patches mjlab classes. make_env() (used by the supplied
scripts) registers the environment automatically. If you construct
ManagerBasedRlEnv yourself and want to retain the historical three-argument
CMAESOptimizer.update_simulator(robot, ...) call, register it explicitly:
env = ManagerBasedRlEnv(cfg=cfg, device="cuda:0")
bind_environment(env)
optimizer.update_simulator(env.scene["robot"], joint_ids, initial_encoder_position)
New applications should prefer the explicit four-argument form with env.