solved_episodes = []
for step in range(args.num_steps):
# render
if args.render:
envs.render()
# select action
value, action, action_log_probs = agent.select_action(step)
# take a step in the environment
obs, reward, done, infos = envs.step(action)
# calculate intrinsic reward
if args.add_intrinsic_reward:
intrinsic_reward = args.intrinsic_coef * agent.compute_intrinsic_reward(
step)
if args.max_intrinsic_reward is not None:
intrinsic_reward = torch.clamp(
agent.compute_intrinsic_reward(step), 0.0,
args.max_intrinsic_reward)
else:
intrinsic_reward = torch.tensor(0).view(1, 1)
intrinsic_rewards.extend(list(
intrinsic_reward.numpy().reshape(-1)))
# store experience
agent.store_rollout(obs[1], action, action_log_probs, value,
reward, intrinsic_reward, done)
# get final episode rewards
for info in infos:
solved_episodes = []
for step in range(args.num_steps):
# render
if args.render:
envs.render()
# select action
value, action, action_log_probs = agent.select_action(step)
# take a step in the environment
obs, reward, done, infos = envs.step(action)
# calculate intrinsic reward
if args.add_intrinsic_reward:
intrinsic_reward = args.intrinsic_coef * agent.compute_intrinsic_reward(step)
if args.max_intrinsic_reward is not None:
intrinsic_reward = torch.clamp(agent.compute_intrinsic_reward(step), 0.0, args.max_intrinsic_reward)
else:
intrinsic_reward = torch.tensor(0).view(1, 1)
intrinsic_rewards.extend(list(intrinsic_reward.numpy().reshape(-1)))
# store experience
agent.store_rollout(obs[1], action, action_log_probs,
value, reward, intrinsic_reward,
done)
# get final episode rewards
for info in infos:
if 'episode' in info.keys():
extrinsic_rewards.append(info['episode']['r'])
