def step(self, action):
observation, reward, done, info = ProxyEnv.step(self, action)
info['reward_inner'] = reward_inner = self.inner_weight * reward
obsIdx = observation[self.idx, ]
info['distance'] = dist = np.linalg.norm(obsIdx - self.current_goal)
info['reward_dist'] = reward_dist = self._compute_dist_reward(obsIdx)
if self.terminal_bonus and dist <= self.terminal_eps:
# print("*****done!!*******")
done = True
reward_dist += self.terminal_bonus
return (self.get_current_obs(), reward_dist + reward_inner, done, info)
def step(self, action):
xy_pos = self.transform_to_goal_space(
self.wrapped_env.get_current_obs())
observation, reward, done, info = ProxyEnv.step(self, action)
info['reward_inner'] = reward_inner = self.inner_weight * reward
# print("REWARD INNER", reward)
if 'distance' not in info:
info['distance'] = dist = self.dist_to_goal(observation)
info['reward_dist'] = reward_dist = self.compute_dist_reward(
observation)
info['goal_reached'] = 1.0 * self.is_goal_reached(observation)
else:
# modified so that inner environment can pass in goal via step
dist = info['distance']
# info['goal_reached'] = 1.0 * (dist < self.terminal_eps)
info['goal_reached'] = 1.0 * self.is_goal_reached(observation)
info[
'reward_dist'] = reward_dist = -self.extend_dist_rew_weight * dist
info['goal'] = self.current_goal
info['obs2goal'] = self._obs2goal_transform(observation)
info['xy_pos'] = xy_pos
info['xy_pos_new'] = self.transform_to_goal_space(observation)
# print(reward_dist)
# print(reward_inner)
# print("step: obs={}, goal={}, dist={}".format(self.append_goal_observation(observation), self.current_goal, dist))
if self.terminate_env and info['goal_reached']:
done = True
if self.append_goal_to_observation:
observation = self.append_goal_observation(observation)
if not self.include_goal_obs:
observation = observation[2:]
return (observation, reward_dist + reward_inner +
info['goal_reached'] * self.goal_weight, done, info)
