robin-shaun / multi-uav-task-assignment-benchmark Goto Github PK

Great work. It is easy to integrate new algorithms and the graphical output is awesome.

I have only 4 minor issues:

1. https://numba.pydata.org/ is easy to use and can speed up GA up to factor 100
   without significant code changes.

2. Only weak algorithms are provided. Very nice for pedagogical purposes, but
   a state-of-the-art algorithm which is challenging to beat is missing.

3. multiprocessing.Pool creates daemonic processes. This prevents experiments
   with multi-threaded algorithms.

4. A multi-objective problem variant - together with the corresponding optimizer(s) is missing.

I created a fork https://github.com/dietmarwo/Multi-UAV-Task-Assignment-Benchmark
fixing all these issues. I can create pull requests if you are interested in some of these fixes. See also https://github.com/dietmarwo/fast-cma-es/blob/master/tutorials/UAV.adoc .

We should make sure that a future comparison with reinforcement learning is fair:
Machine learning uses many GPU cores, so we should utilize parallelization
also when applying optimization.

self.k1 = 0.03 is defined in the file aco.py but not used. What does k1 stand for？

aco_task_assignmet = aco_result.get()[0]

File "D:\Anaconda3\lib\multiprocessing\pool.py", line 657, in get
raise self._value
ValueError: list.remove(x): x not in list
Hello, the above problems may occur in the process of operation. After repeated run, it is found that this problem sometimes occurs early and sometimes late. May I ask why?

运行evaluate.py后，过程中会出现aco.py里面的报错，错误发生在第82行unvisit_list.remove(visit_next)#update，报错内容大致是remove()失败，没有visit_next内容什么的

作者您好，请问在ga.py中gene的前vehicle_num-1个元素代表什么，以及ins的作用是什么，没有想清楚怎样根据seq将任务分配到vehicle_num辆无人机

Could it be possible to adapt all the algorithms to consider the constraint of returning to the depot? How should it be made?