def _step()

in gym/gym/envs/mujoco/humanoid-new.py [0:0]

• 54 lines of code
• 10 McCabe index (conditional complexity)

    def _step(self, a):
        self.timer += 1

        pos_before = mass_center(self.model)
        self.do_simulation(a, self.frame_skip)

        iq = np.copy(self.model.data.qpos)[:,0]
        iv = np.copy(self.model.data.qvel)[:,0]
        iq[-1] = 30
        if self.realgoal == 1:
            iq[-1] = 30
        self.set_state(iq, iv)

        if self.realgoal == 0:
            pos_after = mass_center(self.model)
            alive_bonus = 5.0
            data = self.model.data
            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 = 0 - quad_ctrl_cost - quad_impact_cost
            qpos = self.model.data.qpos
            if not bool((qpos[2] > 0.5)):
                reward += alive_bonus + lin_vel_cost
            done = bool((qpos[2] > 0.5))
            if self.timer < 50:
                done = False
        elif self.realgoal == 1:
            pos_after = mass_center(self.model)
            alive_bonus = 5.0
            data = self.model.data
            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 = 0 - quad_ctrl_cost - quad_impact_cost
            qpos = self.model.data.qpos
            if not bool((qpos[2] < 1.0)):
                reward += alive_bonus + lin_vel_cost
            done = bool((qpos[2] < 1.0))
            done = False
        elif self.realgoal == 2:
            pos_after = mass_center(self.model)
            alive_bonus = 5.0
            data = self.model.data
            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 = 0 - quad_ctrl_cost - quad_impact_cost
            qpos = self.model.data.qpos
            if not bool((qpos[2] < 1.0)):
                reward += alive_bonus - lin_vel_cost
            done = bool((qpos[2] < 1.0))
            done = False


        # pos_after = mass_center(self.model)
        # alive_bonus = 5.0
        # data = self.model.data
        # 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 = 0 - quad_ctrl_cost - quad_impact_cost
        # qpos = self.model.data.qpos
        # if not bool((qpos[2] < 1.0)):
        #     reward += self.model.data.qpos[2]
        # done = bool((qpos[2] < 1.0))

        # print(qpos[2])
        # done = False

        return self._get_obs(), reward, done, {}