Mujoco-py: `TypeError:'numpy.int32' object is not iterable` for make('FetchReach-v1')

[gtatiya]

Describe the bug
I'm able to use env that requires mujoco such as 'CartPole-v1', but can't render any of Fetch robotic env.

To Reproduce

import gym # openAi gym
from gym import envs

env = gym.make('FetchReach-v1')
print(env.reset())
env.render()

Error Messages

TypeError                                 Traceback (most recent call last)
<ipython-input-5-616e18e9e820> in <module>
----> 1 env = gym.make('FetchReach-v1')
      2 print(env.reset())
      3 env.render()

c:\users\gyant\downloads\compressed\gym-master\gym\envs\registration.py in make(id)
    165 
    166 def make(id):
--> 167     return registry.make(id)
    168 
    169 def spec(id):

c:\users\gyant\downloads\compressed\gym-master\gym\envs\registration.py in make(self, id)
    117         logger.info('Making new env: %s', id)
    118         spec = self.spec(id)
--> 119         env = spec.make()
    120         # We used to have people override _reset/_step rather than
    121         # reset/step. Set _gym_disable_underscore_compat = True on

c:\users\gyant\downloads\compressed\gym-master\gym\envs\registration.py in make(self)
     84         else:
     85             cls = load(self._entry_point)
---> 86             env = cls(**self._kwargs)
     87 
     88         # Make the enviroment aware of which spec it came from.

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\fetch\reach.py in __init__(self, reward_type)
     14             gripper_extra_height=0.2, target_in_the_air=True, target_offset=0.0,
     15             obj_range=0.15, target_range=0.15, distance_threshold=0.05,
---> 16             initial_qpos=initial_qpos, reward_type=reward_type)
     17         utils.EzPickle.__init__(self)

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\fetch_env.py in __init__(self, model_path, n_substeps, gripper_extra_height, block_gripper, has_object, target_in_the_air, target_offset, obj_range, target_range, distance_threshold, initial_qpos, reward_type)
     46         super(FetchEnv, self).__init__(
     47             model_path=model_path, n_substeps=n_substeps, n_actions=4,
---> 48             initial_qpos=initial_qpos)
     49 
     50     # GoalEnv methods

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\robot_env.py in __init__(self, model_path, initial_qpos, n_actions, n_substeps)
     32 
     33         self.seed()
---> 34         self._env_setup(initial_qpos=initial_qpos)
     35         self.initial_state = copy.deepcopy(self.sim.get_state())
     36 

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\fetch_env.py in _env_setup(self, initial_qpos)
    170     def _env_setup(self, initial_qpos):
    171         for name, value in initial_qpos.items():
--> 172             self.sim.data.set_joint_qpos(name, value)
    173         utils.reset_mocap_welds(self.sim)
    174         self.sim.forward()

c:\anaconda2\envs\py3_6\lib\site-packages\mujoco_py-1.50.1.68-py3.6.egg\mujoco_py\generated\wrappers.pxi in mujoco_py.cymj.PyMjData.set_joint_qpos()

TypeError: 'numpy.int32' object is not iterable

Desktop (please complete the following information):

• OS: Windows 10
• Python Python 3.6.7
• Mujoco Version 1.50
• mujoco-py version 1.50.1.68

Please Help !!!

[machinaut]

Ah ha! This is a good find! We don't often test these environments on windows, and I think you've found a bug in how we load the environment.

For now can you verify that the Humanoid-v2 environment works for you?

[gtatiya]

Ah ha! This is a good find! We don't often test these environments on windows, and I think you've found a bug in how we load the environment.

For now can you verify that the Humanoid-v2 environment works for you?

Yes, Humanoid-v2 works! But, I got the same error for HandReach-v0.

[machinaut]

if you checkout gym and apply this patch manually, does it work https://github.com/openai/gym/pull/1220 ?

it will probably get merged after some review, but it would be good to test it now.

[gtatiya]

I manually patched for fetch and hand. But, got the same error:

TypeError                                 Traceback (most recent call last)
<ipython-input-4-616e18e9e820> in <module>
----> 1 env = gym.make('FetchReach-v1')
      2 print(env.reset())
      3 env.render()

c:\users\gyant\downloads\compressed\gym-master\gym\envs\registration.py in make(id)
    165 
    166 def make(id):
--> 167     return registry.make(id)
    168 
    169 def spec(id):

c:\users\gyant\downloads\compressed\gym-master\gym\envs\registration.py in make(self, id)
    117         logger.info('Making new env: %s', id)
    118         spec = self.spec(id)
--> 119         env = spec.make()
    120         # We used to have people override _reset/_step rather than
    121         # reset/step. Set _gym_disable_underscore_compat = True on

c:\users\gyant\downloads\compressed\gym-master\gym\envs\registration.py in make(self)
     84         else:
     85             cls = load(self._entry_point)
---> 86             env = cls(**self._kwargs)
     87 
     88         # Make the enviroment aware of which spec it came from.

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\fetch\reach.py in __init__(self, reward_type)
     19             gripper_extra_height=0.2, target_in_the_air=True, target_offset=0.0,
     20             obj_range=0.15, target_range=0.15, distance_threshold=0.05,
---> 21             initial_qpos=initial_qpos, reward_type=reward_type)
     22         utils.EzPickle.__init__(self)

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\fetch_env.py in __init__(self, model_path, n_substeps, gripper_extra_height, block_gripper, has_object, target_in_the_air, target_offset, obj_range, target_range, distance_threshold, initial_qpos, reward_type)
     46         super(FetchEnv, self).__init__(
     47             model_path=model_path, n_substeps=n_substeps, n_actions=4,
---> 48             initial_qpos=initial_qpos)
     49 
     50     # GoalEnv methods

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\robot_env.py in __init__(self, model_path, initial_qpos, n_actions, n_substeps)
     32 
     33         self.seed()
---> 34         self._env_setup(initial_qpos=initial_qpos)
     35         self.initial_state = copy.deepcopy(self.sim.get_state())
     36 

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\fetch_env.py in _env_setup(self, initial_qpos)
    170     def _env_setup(self, initial_qpos):
    171         for name, value in initial_qpos.items():
--> 172             self.sim.data.set_joint_qpos(name, value)
    173         utils.reset_mocap_welds(self.sim)
    174         self.sim.forward()

c:\anaconda2\envs\py3_6\lib\site-packages\mujoco_py-1.50.1.68-py3.6.egg\mujoco_py\generated\wrappers.pxi in mujoco_py.cymj.PyMjData.set_joint_qpos()

TypeError: 'numpy.int32' object is not iterable

[machinaut]

Can you verify that you can load the XML in mujoco (without python) using the simulate binary?

[gtatiya]

Can you verify that you can load the XML in mujoco (without python) using the simulate binary?

Yes, I can run simulate ../model/humanoid.xml

[gtatiya]

Can you verify that you can load the XML in mujoco (without python) using the simulate binary?

Could you please help me.

[machinaut]

@gtatiya can you manually load the XMLs from the fetch and hand environments in mujoco-py?

It would help me if you could provide more context on this error here:

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\fetch_env.py in _env_setup(self, initial_qpos)
    170     def _env_setup(self, initial_qpos):
    171         for name, value in initial_qpos.items():
--> 172             self.sim.data.set_joint_qpos(name, value)
    173         utils.reset_mocap_welds(self.sim)
    174         self.sim.forward()

c:\anaconda2\envs\py3_6\lib\site-packages\mujoco_py-1.50.1.68-py3.6.egg\mujoco_py\generated\wrappers.pxi in mujoco_py.cymj.PyMjData.set_joint_qpos()

TypeError: 'numpy.int32' object is not iterable

In particular, could you provide what name, and value are when that line fails, and provide the model (sim.model)? Thanks

[sino30535]

I just start play with these environments today and ran into same errors, hope these will help.

[sino30535]

So my error message said something like this

File "mujoco_py\generated/wrappers.pxi", line 2547, in mujoco_py.cymj.PyMjData.set_joint_qpos
TypeError: 'numpy.int32' object is not iterable

so I check wrappers.pxi, line 2547,
start_i, end_i = addr

addr was came from line 2535, addr = self._model.get_joint_qpos_addr(name).

The error was caused by addr is a numpy.int32, not a tuple, and when ndim =1, get_joint_qpos_addr return addr instead of (addr, addr+ndim) , which is the reason of this error.

[gtatiya]

@gtatiya can you manually load the XMLs from the fetch and hand environments in mujoco-py?

It would help me if you could provide more context on this error here:

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\fetch_env.py in _env_setup(self, initial_qpos)
    170     def _env_setup(self, initial_qpos):
    171         for name, value in initial_qpos.items():
--> 172             self.sim.data.set_joint_qpos(name, value)
    173         utils.reset_mocap_welds(self.sim)
    174         self.sim.forward()

c:\anaconda2\envs\py3_6\lib\site-packages\mujoco_py-1.50.1.68-py3.6.egg\mujoco_py\generated\wrappers.pxi in mujoco_py.cymj.PyMjData.set_joint_qpos()

TypeError: 'numpy.int32' object is not iterable

In particular, could you provide what name, and value are when that line fails, and provide the model (sim.model)? Thanks

Yes, simulate reach.xml works.
Here's the output you asked:

name:  robot0:slide0
value:  0.4049
self.sim.model:  <mujoco_py.cymj.PyMjModel object at 0x000001A7145DF908>

[xiaodaxia-2008]

i've gone to the same problem...

[gtatiya]

@gtatiya can you manually load the XMLs from the fetch and hand environments in mujoco-py?

It would help me if you could provide more context on this error here:

c:\users\gyant\downloads\compressed\gym-master\gym\envs\robotics\fetch_env.py in _env_setup(self, initial_qpos)
    170     def _env_setup(self, initial_qpos):
    171         for name, value in initial_qpos.items():
--> 172             self.sim.data.set_joint_qpos(name, value)
    173         utils.reset_mocap_welds(self.sim)
    174         self.sim.forward()

c:\anaconda2\envs\py3_6\lib\site-packages\mujoco_py-1.50.1.68-py3.6.egg\mujoco_py\generated\wrappers.pxi in mujoco_py.cymj.PyMjData.set_joint_qpos()

TypeError: 'numpy.int32' object is not iterable

In particular, could you provide what name, and value are when that line fails, and provide the model (sim.model)? Thanks

Could you please help me.

[machinaut]

@gtatiya I've not been able to reproduce anything like this, so I still am confused as to the source of the error.

Do you know what the numpy.int32 object in question is?

[sino30535]

@machinaut
in wrappers.pxi, line 1050 - 1052

    Returns:
    - address (int, tuple): returns int address if 1-dim joint, otherwise
        returns the a (start, end) tuple for pos[start:end] access.

I think addr is the numpy.int32 object that causes error.

[Marc-Ducret]

I ran into the same issue.
I experimented with wrappers.pxi.

in set_joint_qpos(self, name, value) (line 2542):
print(type(addr), isinstance(addr, (int, np.int32, np.int64))) prints <class 'numpy.int32'> False.

Replacing all instances of isinstance(addr, (int, np.int32, np.int64))) with hasattr(addr, '__int__') seems to fix the issue. I am however not sure how wrappers.pxi is generated and what should be the proper change to fix the issue.

[machinaut]

@VengeurK There's a script (probably easy to find) which generates this, https://github.com/openai/mujoco-py/blob/master/scripts/gen_wrappers.py

[aaronsnoswell]

I can confirm @VengeurK's fix works for me too, I'm preparing a PR that fixes the windows build.

[Altriaex]

I still encounter the same bug. Any update now?

[aaronsnoswell]

@Altriaex Are you able to try my branch referenced in the pull request above? OpenAI hasn't responded to the PR, but it may fix this issue for you.

[Altriaex]

@aaronsnoswell
It turns out that, I need to recompile it, after editing files, as you mentioned in
https://github.com/openai/mujoco-py/compare/master...aaronsnoswell:fix-windows-support

Thank you!

[YangRui2015]

I met the same problem using mujoco_py on Windows, but the lines I need to change were around 1054, and there were about 4 places to replace.
My mujoco_py version is 1.50.1.0.