def _init():
register_gripper(UltrasoundProbeGripper)
env = GymWrapper(suite.make(env_id, **options))
env = Monitor(env)
env.seed(seed + rank)
return env
model = PPO.load(continue_training_model_path, env=env)
# Training
model.learn(total_timesteps=training_timesteps,
tb_log_name=tb_log_name,
callback=checkpoint_callback,
reset_num_timesteps=True)
# Save trained model
model.save(save_model_path)
env.save(save_vecnormalize_path)
else:
# Create evaluation environment
env_options['has_renderer'] = True
register_gripper(UltrasoundProbeGripper)
env_gym = GymWrapper(suite.make(env_id, **env_options))
env = DummyVecEnv([lambda: env_gym])
# Load normalized env
env = VecNormalize.load(load_vecnormalize_path, env)
# Turn of updates and reward normalization
env.training = False
env.norm_reward = False
# Load model
model = PPO.load(load_model_path, env)
# Simulate environment
obs = env.reset()