I have a python class which communicates via a ROS service with a simulation environment. Now I want to communicate with multiple simulations in parallel.
In detail, the environment class has a step function which takes an argument and returns the simulation observations:
self.envs = [Environment("AI1"),
Environment("AI2")]
self.pool = Pool(processes=len(self.envs))
ac = # some actions
res0 = self.pool.apply_async(self.envs[0].step, ac[0])
res1 = self.pool.apply_async(self.envs[1].step, ac[1])
res_list = []
res_list.extend(res0.get())
res_list.extend(res1.get())
Of course, I get now the following error
cPickle.PicklingError: Can't pickle : attribute lookup builtin.instancemethod failed
because a ROSPy service is a member of Environment
Is it possible to create the environments in the pool itself? Or would switching to python threading help?
With threads you won't get this issue.
An alternative is to start subprocesses where the subprocess creates the Environment and listens on a queue for requests and pushes responses onto a different queue. That way the environment lives in the subprocess.