dodox/fancy_gym
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

minor bug fixes

Browse Source
This commit is contained in:
Fabian 2022-07-26 10:33:59 +02:00
parent ce795669a9
commit 4aacd71ed3
2 changed files with 26 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
fancy_gym/black_box/black_box_wrapper.py
View File

@ -75,19 +75,24 @@ class BlackBoxWrapper(gym.ObservationWrapper):
clipped_params = np.clip(action, self.traj_gen_action_space.low, self.traj_gen_action_space.high)
self.traj_gen.set_params(clipped_params)
# TODO: is this correct for replanning? Do we need to adjust anything here?
self.traj_gen.set_boundary_conditions(
bc_time=np.array(0) if not self.do_replanning else np.array([self.current_traj_steps * self.dt]),
bc_pos=self.current_pos, bc_vel=self.current_vel)
# TODO remove the - self.dt after Bruces fix.
bc_time = np.array(0 if not self.do_replanning else self.current_traj_steps * self.dt)
self.traj_gen.set_boundary_conditions(bc_time, self.current_pos, self.current_vel)
# TODO: remove the - self.dt after Bruces fix.
self.traj_gen.set_duration(None if self.learn_sub_trajectories else self.duration - self.dt, self.dt)
traj_dict = self.traj_gen.get_trajs(get_pos=True, get_vel=True)
trajectory_tensor, velocity_tensor = traj_dict['pos'], traj_dict['vel']
# traj_dict = self.traj_gen.get_trajs(get_pos=True, get_vel=True)
trajectory = get_numpy(self.traj_gen.get_traj_pos())
velocity = get_numpy(self.traj_gen.get_traj_vel())
return get_numpy(trajectory_tensor), get_numpy(velocity_tensor)
if self.do_replanning:
# Remove first part of trajectory as this is already over
trajectory = trajectory[self.current_traj_steps:]
velocity = velocity[self.current_traj_steps:]
return trajectory, velocity
def _get_traj_gen_action_space(self):
"""This function can be used to set up an individual space for the parameters of the traj_gen."""
min_action_bounds, max_action_bounds = self.traj_gen.get_params_bounds().t()
min_action_bounds, max_action_bounds = self.traj_gen.get_params_bounds()
action_space = gym.spaces.Box(low=min_action_bounds.numpy(), high=max_action_bounds.numpy(),
dtype=self.env.action_space.dtype)
return action_space
@ -105,13 +110,13 @@ class BlackBoxWrapper(gym.ObservationWrapper):
return self._get_traj_gen_action_space()
def _get_observation_space(self):
if self.return_context_observation:
mask = self.env.context_mask
if not self.return_context_observation:
# return full observation
mask = np.ones_like(mask, dtype=bool)
min_obs_bound = self.env.observation_space.low[mask]
max_obs_bound = self.env.observation_space.high[mask]
return spaces.Box(low=min_obs_bound, high=max_obs_bound, dtype=self.env.observation_space.dtype)
return self.env.observation_space
def step(self, action: np.ndarray):
""" This function generates a trajectory based on a MP and then does the usual loop over reset and step"""
@ -152,18 +157,18 @@ class BlackBoxWrapper(gym.ObservationWrapper):
t + 1 + self.current_traj_steps):
break
infos.update({k: v[:t + 1] for k, v in infos.items()})
infos.update({k: v[:t] for k, v in infos.items()})
self.current_traj_steps += t + 1
if self.verbose >= 2:
infos['positions'] = trajectory
infos['velocities'] = velocity
infos['step_actions'] = actions[:t]
infos['step_actions'] = actions[:t + 1]
infos['step_observations'] = observations[:t + 1]
infos['step_rewards'] = rewards[:t]
infos['step_rewards'] = rewards[:t + 1]
infos['trajectory_length'] = t + 1
trajectory_return = self.reward_aggregation(rewards[:t])
trajectory_return = self.reward_aggregation(rewards[:t + 1])
return self.observation(obs), trajectory_return, done, infos
def render(self, **kwargs):

5
fancy_gym/utils/time_aware_observation.py
View File

@ -40,9 +40,10 @@ class TimeAwareObservation(gym.ObservationWrapper):
high = np.append(self.observation_space.high, 1.0)
self.observation_space = Box(low, high, dtype=self.observation_space.dtype)
self.t = 0
self._max_episode_steps = env.spec.max_episode_steps
def observation(self, observation):
"""Adds to the observation with the current time step.
"""Adds to the observation with the current time step normalized with max steps.
Args:
observation: The observation to add the time step to
@ -50,7 +51,7 @@ class TimeAwareObservation(gym.ObservationWrapper):
Returns:
The observation with the time step appended to
"""
return np.append(observation, self.t/self.env.spec.max_episode_steps)
return np.append(observation, self.t / self._max_episode_steps)
def step(self, action):
"""Steps through the environment, incrementing the time step.