ImportError: No module named joint0[0,0,0]_Tx.wrapper_module_0
cambel opened this issue · 3 comments
Hi there,
I have been following your work recently (Play around with the sympy lib and a UR3 arm). Now I was trying to use this abr_control library, I made it work with the UR3 model and the first time it runs kind of fine. I take about 10 minutes computing every... kinda slow but anyway it works. But when I try to run it again I got an error about loading a saved function. I tried with the pygame example to see if it was not a problem with my custom model, but I got the same thing after trying to run the example a 2nd time:
Traceback (most recent call last): File "reaching_sliding.py", line 30, in <module> interface.connect() File "/home/cambel/dev/abr_control/abr_control/interfaces/ipygame.py", line 87, in connect self.arm_sim.connect() File "/home/cambel/dev/abr_control/abr_control/arms/threejoint/arm_sim.py", line 54, in connect self._update_state() File "/home/cambel/dev/abr_control/abr_control/arms/threejoint/arm_sim.py", line 121, in _update_state self._position() File "/home/cambel/dev/abr_control/abr_control/arms/threejoint/arm_sim.py", line 109, in _position for ii in range(self.robot_config.N_JOINTS)] File "/home/cambel/dev/abr_control/abr_control/arms/base_config.py", line 362, in Tx self._Tx[funcname] = self._calc_Tx(name, x=x) File "/home/cambel/dev/abr_control/abr_control/arms/base_config.py", line 754, in _calc_Tx Tx, Tx_func = self._load_from_file(filename, lambdify) File "/home/cambel/dev/abr_control/abr_control/arms/base_config.py", line 188, in _load_from_file filename + '.' + saved_file) File "/usr/lib/python2.7/importlib/__init__.py", line 37, in import_module __import__(name) ImportError: No module named joint0[0,0,0]_Tx.wrapper_module_0
It seems to be unable to find the saved function.
Hello!
In the example script, when you create the robot_config = Config(use_cython=True) if you change use_cython=False you should see significantly faster start up (because it won't be generating C functions for each of the calculations), but the execution time may be a bit slower. Probably that's fine though.
I believe that the import error is due to using Python 2.7. If you download and setup Anaconda, then it should be very quick to get set up in Python 3 running conda create -n myenv python=3.5. From there it should be quick to try running the examples and hopefully when you use use_cython=True there aren't any issues. Actually, I find that when I'm using Python 2.7 I'm not even able to run the example a first time without an import error. I will make a note that Python 2 isn't supported, thanks!
Regarding building up a config for the UR3, awesome! Would love to get your feedback on the process, and if you have any questions or we can help with it please don't hesitate to write.
So, tl;dr: I think the import errors are cause by Python 2.7, if you switch to Python 3+ those should go away. Alternatively, setting use_cython=False in the scripts your running when creating robot_config might also do the trick. Please let me know!
Yeah, since I was also playing with your blog's sympy example I disable the use_cython option, and it worked fine, thank you very much. I've been using Python 2.7 since I want to use another library that is working on that version.
Now, I have one comment. I wonder why you didn't just save the lambdify function directly using cloudpickle. I tried, and it works fine, both in the sympy example and in the abr_control, it runs faster than loading just the expression and then lambdify it. Not sure if it is quicker or slower in run time than the cython version.
Regarding the building the UR3 model, I had done a couple of things. First, I tried to define joints and links transformations using the DH parameters, and so I implemented some method to computed that on the sympy example. I think it looks kinda simpler to read, and I used the parameter values provided by universal robots for the ur3. After finding the correct offset angles for theta, it worked perfectly with the VREP model. They also provide the values of the CoMs of each link so I also tried to use that in my model and it seems to work fine too.
I haven't tried this in the abr_control library, there I just change the values of the link/joints offsets, the inertia values, and then used the same transformations as the UR5 model.
So, anyway, thank you very much for your blog and your code, it has been really useful to understand a bit more about control. I will keep using it, I actually want to create a torque/force controller with velocity inputs. That's my goal, and then see if it works in the real robot. If you find useful any of my modifications, let me know, I'd be glad to contribute.
Sorry about the crazy long response time, thanks for the example! We've discussed getting a DH parameters setup going for the repo, but are just short on cycles to develop it. If you are able to make a PR for it we would definitely appreciate it as it's super worthwhile to have!