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

use desired point as boundary condition

Browse Source
This commit is contained in:
Hongyi Zhou
2022-10-25 22:15:30 +02:00
parent 556bfd0b35
commit a1d96e6016
3 changed files with 61 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fancy_gym/black_box/black_box_wrapper.py
+22 -4
View File
@@ -21,7 +21,8 @@ class BlackBoxWrapper(gym.ObservationWrapper):
learn_sub_trajectories: bool = False,
replanning_schedule: Optional[
Callable[[np.ndarray, np.ndarray, np.ndarray, np.ndarray, int], bool]] = None,
reward_aggregation: Callable[[np.ndarray], float] = np.sum
reward_aggregation: Callable[[np.ndarray], float] = np.sum,
desired_conditioning: bool = False
):
"""
gym.Wrapper for leveraging a black box approach with a trajectory generator.
@@ -67,6 +68,11 @@ class BlackBoxWrapper(gym.ObservationWrapper):
self.render_kwargs = {}
self.verbose = verbose
# condition value
self.desired_conditioning = True
self.condition_pos = None
self.condition_vel = None
def observation(self, observation):
# return context space if we are
if self.return_context_observation:
@@ -87,7 +93,11 @@ class BlackBoxWrapper(gym.ObservationWrapper):
bc_time = np.array(0 if not self.do_replanning else self.current_traj_steps * self.dt)
# TODO we could think about initializing with the previous desired value in order to have a smooth transition
# at least from the planning point of view.
self.traj_gen.set_boundary_conditions(bc_time, self.current_pos, self.current_vel)
# self.traj_gen.set_boundary_conditions(bc_time, self.current_pos, self.current_vel)
if self.current_traj_steps == 0:
self.condition_pos = self.current_pos
self.condition_vel = self.current_vel
self.traj_gen.set_boundary_conditions(bc_time, self.condition_pos, self.condition_vel)
self.traj_gen.set_duration(duration, self.dt)
# traj_dict = self.traj_gen.get_trajs(get_pos=True, get_vel=True)
position = get_numpy(self.traj_gen.get_traj_pos())
@@ -165,14 +175,22 @@ class BlackBoxWrapper(gym.ObservationWrapper):
if done or self.replanning_schedule(self.current_pos, self.current_vel, obs, c_action,
t + 1 + self.current_traj_steps):
if self.desired_conditioning:
self.condition_pos = pos
self.condition_vel = vel
else:
self.condition_pos = self.current_pos
self.condition_vel = self.current_vel
break
infos.update({k: v[:t+1] for k, v in infos.items()})
self.current_traj_steps += t + 1
if self.verbose >= 2:
infos['positions'] = position
infos['velocities'] = velocity
infos['desired_pos'] = position[:t+1]
infos['desired_vel'] = velocity[:t+1]
infos['current_pos'] = self.current_pos
infos['current_vel'] = self.current_vel
infos['step_actions'] = actions[:t + 1]
infos['step_observations'] = observations[:t + 1]
infos['step_rewards'] = rewards[:t + 1]