def eval_one_episode(TASK_ENV, model_client): model_client.call(func_name="reset") # reset policy while not TASK_ENV.is_episode_end(): # Check whether the episode ends obs = TASK_ENV.get_obs() # Get Observation model_client.call(func_name="update_obs", obs=obs) # Update Observation actions = model_client.call(func_name="get_action") # Get Action according to observation chunk for action_idx, action in enumerate(actions): TASK_ENV.take_action(action) if TASK_ENV.is_episode_end() or action_idx + 1 == len(actions): break obs = TASK_ENV.get_obs() model_client.call(func_name="update_obs", obs=obs) def eval_one_episode_batch(TASK_ENV, model_client): model_client.call(func_name="reset") while not TASK_ENV.is_episode_end(): # Check whether the episode ends env_idx_list = TASK_ENV.get_running_env_idx_list() # Get Running Environment Index List obs_list = TASK_ENV.get_obs_batch(env_idx_list) # Get Observation model_client.call(func_name="update_obs_batch", obs=obs_list) actions = model_client.call(func_name="get_action_batch", obs=env_idx_list) # Get Action according to observation chunk chunk_size = len(actions[0]) # Get the chunk size for action_idx in range(chunk_size): # Iterate over the action chunk current_action_list = [env_actions[action_idx] for env_actions in actions] # Get the current action list TASK_ENV.take_action_batch(current_action_list, env_idx_list) # Take the action if TASK_ENV.is_episode_end() or action_idx + 1 == chunk_size: # Check whether the episode ends break running = set(TASK_ENV.get_running_env_idx_list()) # Get the running environment index list active_batch_idx = [i for i, env_idx in enumerate(env_idx_list) if env_idx in running] # Get the active batch index actions = [actions[i] for i in active_batch_idx] # Get the active action list env_idx_list = [env_idx_list[i] for i in active_batch_idx] # Get the active environment index list model_client.call(func_name="update_obs_batch", obs=TASK_ENV.get_obs_batch(env_idx_list)) # Update the observation