def step(self, a):
pos_before = mass_center(self.model, self.sim)
self.do_simulation(a, self.frame_skip)
pos_after = mass_center(self.model, self.sim)
pos_after_standup = self.sim.data.qpos[2]
down = bool((pos_after_standup < 1.0) or (pos_after_standup > 2.0))
alive_bonus = 5.0 if not down else 1.0
data = self.sim.data
uph_cost = (pos_after_standup - 0) / self.model.opt.timestep
lin_vel_cost = 0.25 * (pos_after -
pos_before) / self.model.opt.timestep
quad_ctrl_cost = 0.1 * np.square(data.ctrl).sum()
quad_impact_cost = .5e-6 * np.square(data.cfrc_ext).sum()
quad_impact_cost = min(quad_impact_cost, 10)
reward = lin_vel_cost + uph_cost - quad_ctrl_cost - quad_impact_cost + alive_bonus
done = bool(False)
return self._get_obs(), reward, done, dict(
reward_linup=uph_cost,
reward_quadctrl=-quad_ctrl_cost,
reward_impact=-quad_impact_cost,
reward_alive=alive_bonus)
def _step(self, a):
pos_before = mass_center(self.model)
self.do_simulation(a, self.frame_skip)
pos_after = mass_center(self.model)
pos_after_standup = self.model.data.qpos[2][0]
down = bool(( self.model.data.qpos[2] < 1.0) or ( self.model.data.qpos[2] > 2.0))
alive_bonus = 5.0 if not down else 1.0
data = self.model.data
uph_cost = (pos_after_standup - 0) / self.model.opt.timestep
lin_vel_cost = 0.25 * (pos_after - pos_before) / self.model.opt.timestep
quad_ctrl_cost = 0.1 * np.square(data.ctrl).sum()
quad_impact_cost = .5e-6 * np.square(data.cfrc_ext).sum()
quad_impact_cost = min(quad_impact_cost, 10)
reward = lin_vel_cost + uph_cost - quad_ctrl_cost - quad_impact_cost + alive_bonus
qpos = self.model.data.qpos
done = bool(False)
return self._get_obs(), reward, done, dict(reward_linup=uph_cost, reward_quadctrl=-quad_ctrl_cost, reward_impact=-quad_impact_cost, reward_alive=alive_bonus)
def step(self, a):
# a = self._add_noise(a)
ob, reward, done, info = super().step(a)
mass_center_after = mass_center(self.model, self.sim)
if mass_center_after > self._base_pos:
self._base_pos += 1.
sparse_rew = 1
else:
sparse_rew = 0
info['original_rew'] = reward
return ob, sparse_rew, done, info
def step(self, a):
pos_before = self.sim.data.qpos[2]
self.do_simulation(a, self.frame_skip)
pos_after = self.sim.data.qpos[2]
data = self.sim.data
uph_cost = (pos_after - 0) / self.model.opt.timestep
up_vel = (pos_after - pos_before) / self.model.opt.timestep
quad_ctrl_cost = 0.1 * np.square(data.ctrl).sum()
quad_impact_cost = .5e-6 * np.square(data.cfrc_ext).sum()
quad_impact_cost = min(quad_impact_cost, 10)
reward = uph_cost - quad_ctrl_cost - quad_impact_cost + 1
#print(reward)
subtask_1_reward = mass_center(self.model, self.sim)
#print('subtask_1_reward', mass_center(self.model, self.sim))
subtask_2_reward = up_vel
done = bool(False)
return self._get_obs(), reward, done, \
dict(reward_linup=uph_cost,
reward_quadctrl=-quad_ctrl_cost,
reward_impact=-quad_impact_cost,
subtask_1 = subtask_1_reward,
subtask_2 = subtask_2_reward)
def _get_obs(self):
data = self.sim.data
center = mass_center(self.model, self.sim)
obs = super()._get_obs()
return np.concatenate([obs, [center]])