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

some changes to reward

Browse Source
This commit is contained in:
ottofabian 2020-08-31 15:51:47 +02:00
parent 7f4f52ab10
commit 3cc4d6e667
2 changed files with 10 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
alr_envs/__init__.py
View File

@ -11,6 +11,6 @@ register(
entry_point='alr_envs.classic_control:SimpleReacherEnv', entry_point='alr_envs.classic_control:SimpleReacherEnv',
max_episode_steps=200, max_episode_steps=200,
kwargs={ kwargs={
"n_links": 5, "n_links": 3,
} }
) )

16
alr_envs/classic_control/simple_reacher.py
View File

@ -61,12 +61,12 @@ class SimpleReacherEnv(gym.Env, utils.EzPickle):
self._update_joints() self._update_joints()
self._steps += 1 self._steps += 1
reward = self._get_reward(action) reward, info = self._get_reward(action)
# done = np.abs(self.end_effector - self._goal_pos) < 0.1 # done = np.abs(self.end_effector - self._goal_pos) < 0.1
done = False done = False
return self._get_obs().copy(), reward, done, {} return self._get_obs().copy(), reward, done, info
def _scale_action(self, action): def _scale_action(self, action):
""" """
@ -107,15 +107,17 @@ class SimpleReacherEnv(gym.Env, utils.EzPickle):
def _get_reward(self, action): def _get_reward(self, action):
diff = self.end_effector - self._goal_pos diff = self.end_effector - self._goal_pos
distance = 0 reward_dist = 0
# TODO: Is this the best option # TODO: Is this the best option
if self._steps >= self.steps_before_reward: if self._steps >= self.steps_before_reward:
distance = np.exp(-0.1 * diff ** 2).mean() reward_dist = - np.linalg.norm(diff)
# distance -= (diff ** 2).mean() # reward_dist = np.exp(-0.1 * diff ** 2).mean()
# reward_dist = - (diff ** 2).mean()
# distance -= action ** 2 reward_ctrl = (action ** 2).sum()
return distance reward = reward_dist - reward_ctrl
return reward, dict(reward_dist=reward_dist, reward_ctrl=reward_ctrl)
def reset(self): def reset(self):