gym/envs/mujoco/pusher.py [22:77]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
            self, "pusher.xml", 5, observation_space=observation_space, **kwargs
        )

    def step(self, a):
        vec_1 = self.get_body_com("object") - self.get_body_com("tips_arm")
        vec_2 = self.get_body_com("object") - self.get_body_com("goal")

        reward_near = -np.linalg.norm(vec_1)
        reward_dist = -np.linalg.norm(vec_2)
        reward_ctrl = -np.square(a).sum()
        reward = reward_dist + 0.1 * reward_ctrl + 0.5 * reward_near

        self.do_simulation(a, self.frame_skip)
        if self.render_mode == "human":
            self.render()

        ob = self._get_obs()
        return (
            ob,
            reward,
            False,
            False,
            dict(reward_dist=reward_dist, reward_ctrl=reward_ctrl),
        )

    def viewer_setup(self):
        assert self.viewer is not None
        self.viewer.cam.trackbodyid = -1
        self.viewer.cam.distance = 4.0

    def reset_model(self):
        qpos = self.init_qpos

        self.goal_pos = np.asarray([0, 0])
        while True:
            self.cylinder_pos = np.concatenate(
                [
                    self.np_random.uniform(low=-0.3, high=0, size=1),
                    self.np_random.uniform(low=-0.2, high=0.2, size=1),
                ]
            )
            if np.linalg.norm(self.cylinder_pos - self.goal_pos) > 0.17:
                break

        qpos[-4:-2] = self.cylinder_pos
        qpos[-2:] = self.goal_pos
        qvel = self.init_qvel + self.np_random.uniform(
            low=-0.005, high=0.005, size=self.model.nv
        )
        qvel[-4:] = 0
        self.set_state(qpos, qvel)
        return self._get_obs()

    def _get_obs(self):
        return np.concatenate(
            [
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



gym/envs/mujoco/pusher_v4.py [144:199]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
            self, "pusher.xml", 5, observation_space=observation_space, **kwargs
        )

    def step(self, a):
        vec_1 = self.get_body_com("object") - self.get_body_com("tips_arm")
        vec_2 = self.get_body_com("object") - self.get_body_com("goal")

        reward_near = -np.linalg.norm(vec_1)
        reward_dist = -np.linalg.norm(vec_2)
        reward_ctrl = -np.square(a).sum()
        reward = reward_dist + 0.1 * reward_ctrl + 0.5 * reward_near

        self.do_simulation(a, self.frame_skip)
        if self.render_mode == "human":
            self.render()

        ob = self._get_obs()
        return (
            ob,
            reward,
            False,
            False,
            dict(reward_dist=reward_dist, reward_ctrl=reward_ctrl),
        )

    def viewer_setup(self):
        assert self.viewer is not None
        self.viewer.cam.trackbodyid = -1
        self.viewer.cam.distance = 4.0

    def reset_model(self):
        qpos = self.init_qpos

        self.goal_pos = np.asarray([0, 0])
        while True:
            self.cylinder_pos = np.concatenate(
                [
                    self.np_random.uniform(low=-0.3, high=0, size=1),
                    self.np_random.uniform(low=-0.2, high=0.2, size=1),
                ]
            )
            if np.linalg.norm(self.cylinder_pos - self.goal_pos) > 0.17:
                break

        qpos[-4:-2] = self.cylinder_pos
        qpos[-2:] = self.goal_pos
        qvel = self.init_qvel + self.np_random.uniform(
            low=-0.005, high=0.005, size=self.model.nv
        )
        qvel[-4:] = 0
        self.set_state(qpos, qvel)
        return self._get_obs()

    def _get_obs(self):
        return np.concatenate(
            [
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -