rohanpsingh/LearningHumanoidWalking

bug in 'jvrc_walk' env

Opened this issue · 2 comments

Traceback (most recent call last):
  File "E:/3_code/4_lab/3.规划/10_具身智能/6_RA/Codeset/Mujoco/LearningHumanoidWalking-main/run_experiment.py", line 133, in <module>
    run_experiment(args)
  File "E:/3_code/4_lab/3.规划/10_具身智能/6_RA/Codeset/Mujoco/LearningHumanoidWalking-main/run_experiment.py", line 83, in run_experiment
    procs=args.num_procs))
  File "E:\3_code\4_lab\3.规划\10_具身智能\6_RA\Codeset\Mujoco\LearningHumanoidWalking-main\rl\envs\normalize.py", line 71, in get_normalization_params
    states.extend(ray.get(ready_ids[0]))
  File "D:\4_code\1_anaconda\envs\walk\lib\site-packages\ray\_private\client_mode_hook.py", line 105, in wrapper
    return func(*args, **kwargs)
  File "D:\4_code\1_anaconda\envs\walk\lib\site-packages\ray\worker.py", line 1713, in get
    raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(AttributeError): ray::_run_random_actions() (pid=35092, ip=127.0.0.1)
  File "python\ray\_raylet.pyx", line 625, in ray._raylet.execute_task
  File "python\ray\_raylet.pyx", line 629, in ray._raylet.execute_task
  File "E:\3_code\4_lab\3.规划\10_具身智能\6_RA\Codeset\Mujoco\LearningHumanoidWalking-main\rl\envs\normalize.py", line 36, in _run_random_actions
    state, _, done, _ = env.step(action.data.numpy())
  File "E:\3_code\4_lab\3.规划\10_具身智能\6_RA\Codeset\Mujoco\LearningHumanoidWalking-main\rl\envs\wrappers.py", line 13, in step
    state, reward, done, info = self.env.step(action[0])
  File "E:\3_code\4_lab\3.规划\10_具身智能\6_RA\Codeset\Mujoco\LearningHumanoidWalking-main\envs\jvrc\jvrc_walk.py", line 118, in step
    rewards = self.task.calc_reward(self.robot.prev_torque, self.robot.prev_action, applied_action)
  File "E:\3_code\4_lab\3.规划\10_具身智能\6_RA\Codeset\Mujoco\LearningHumanoidWalking-main\tasks\walking_task.py", line 55, in calc_reward
    action_penalty=0.050 * rewards._calc_action_reward(self, prev_action),
  File "E:\3_code\4_lab\3.规划\10_具身智能\6_RA\Codeset\Mujoco\LearningHumanoidWalking-main\tasks\rewards.py", line 17, in _calc_action_reward
    action = self._client.get_pd_target()[0]
AttributeError: 'RobotInterface' object has no attribute 'get_pd_target'

Then i check the 'rewards.py' and 'robot_interface.py' files, and i exactly find that there is no function named get_pd_target. So for trainning, I just comment out the code.But I do't know whether it will have a impact on trained models.

@formoree found in fork of this repo: https://github.com/mhming/LearningHumanoidWalking/blob/main/envs/common/robot_interface.py

adding this code or substituting this code in downloaded repo- files- /LearningHumanoidWalking/envs/common/robot_interface.py

I added a snippet so you know where to add. I can create pull request if it didn't work for you

def check_self_collisions(self):
        """
        Returns True if there are collisions other than any-geom-floor.
        """
        contacts = [self.data.contact[i] for i in range(self.data.ncon)]
        floor_contacts = []
        floor_id = mujoco.mj_name2id(self.model, mujoco.mjtObj.mjOBJ_BODY, self.floor_body_name)
        for i,c in enumerate(contacts):
            geom1_is_floor = (self.model.geom_bodyid[c.geom1]==floor_id)
            geom2_is_floor = (self.model.geom_bodyid[c.geom2]==floor_id)
            if (geom1_is_floor or geom2_is_floor):
                floor_contacts.append((i,c))
        return len(floor_contacts) != self.data.ncon

    def get_pd_target(self):  ## this line is added
        return [self.current_pos_target, self.current_vel_target]  ## this line is added

    def set_pd_gains(self, kp, kv):
        assert kp.size==self.model.nu
        assert kv.size==self.model.nu
        self.kp = kp.copy()
        self.kv = kv.copy()
        return

    def step_pd(self, p, v):
        self.current_pos_target = p.copy()  ## this line is added
        self.current_vel_target = v.copy()  ## this line is added
        target_angles = self.current_pos_target
        target_speeds = self.current_vel_target

        assert type(target_angles)==np.ndarray
        assert type(target_speeds)==np.ndarray

        curr_angles = self.get_act_joint_positions()
        curr_speeds = self.get_act_joint_velocities()

        perror = target_angles - curr_angles
        verror = target_speeds - curr_speeds

        assert self.kp.size==perror.size
        assert self.kv.size==verror.size
        assert perror.size==verror.size
        return self.kp * perror + self.kv * verror

OK. Let me have a try. I think the Uncomment the code, and the code can be runed. But I don't know whether it will have a impact on the tasks.