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

update

Browse Source
This commit is contained in:
Maximilian Huettenrauch
2021-06-22 10:27:25 +02:00
parent 3876478b96
commit 8075655301
3 changed files with 12 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
alr_envs/mujoco/ball_in_a_cup/ball_in_a_cup.py
+5 -1
View File
@@ -113,7 +113,11 @@ class ALRBallInACupEnv(alr_mujoco_env.AlrMujocoEnv, utils.EzPickle):
return ob, reward, done, dict(reward_dist=reward_dist,
reward_ctrl=reward_ctrl,
velocity=angular_vel,
traj=self._q_pos, is_success=success,
# traj=self._q_pos,
action=a,
q_pos=self.sim.data.qpos[0:7].ravel().copy(),
q_vel=self.sim.data.qvel[0:7].ravel().copy(),
is_success=success,
is_collided=is_collided, sim_crash=crash)
def check_traj_in_joint_limits(self):
alr_envs/mujoco/ball_in_a_cup/ball_in_a_cup_reward_simple.py
+2 -2
View File
@@ -81,11 +81,11 @@ class BallInACupReward(alr_reward_fct.AlrReward):
# cost = 0.5 * dist_final + 0.05 * cost_angle # TODO: Increase cost_angle weight # 0.5 * min_dist +
# reward = np.exp(-2 * cost) - 1e-2 * action_cost - self.collision_penalty * int(self._is_collided)
# reward = - dist_final**2 - 1e-4 * cost_angle - 1e-5 * action_cost - self.collision_penalty * int(self._is_collided)
reward = - dist_final**2 - min_dist_final**2 - 1e-4 * cost_angle - 5e-4 * action_cost - self.collision_penalty * int(self._is_collided)
reward = - dist_final**2 - min_dist_final**2 - 1e-4 * cost_angle - 1e-3 * action_cost - self.collision_penalty * int(self._is_collided)
success = dist_final < 0.05 and ball_in_cup and not self._is_collided
crash = self._is_collided
else:
reward = - 5e-4 * action_cost - 1e-4 * cost_angle # TODO: increase action_cost weight
reward = - 1e-3 * action_cost - 1e-4 * cost_angle # TODO: increase action_cost weight
success = False
crash = False