I am not sure, but maybe you can have a look at the compute room on the ground floor of Informatic Department?
I had a look, only 5% of the time is spent on collision detection. Everything else is trajectory feasibility checking together with optimization, implemented in python. The actual time on a powerful machine is given here.
It could be possible to go down to 5+5 sec per scenario configuring search algorithm parameters.
State-of-the-art algorithms have close-to-realtime performance. Search is search.
The compute room on the ground floor has very slow thin clients.
Use Google Colaboratory instead - limited to 12 hours or less, but powerful CPUs. You can run several instances in parallel. Configure once and upload binaries to git - or use Docker.
So how do we install it on colab? Are there any steps? Bcos is not just a normal installation
Hi, When I run gbfs_only_time, I get 0 solutions with 120 timeout. I latest pulled the on the weekend. Any idea what could go wrong or how to run that particular .py file ?
Hi, I just checked again. It should be working.
Here is what I have done:
- download the virtual machine image
- change the planner id in batch processing configuration file to 3
- change MotionPlanner_gbfs_only_time.py to MotionPlanner.py (you can also change the code in batch_search.py to read MotionPlanner_gbfs_only_time.py instead)
- run batch processing
- get results
If it still doesn’t work for you, you might try increasing the timeout time to see if the situation change.
It works. Thanks! However, @jkljkljkl , the same motion_planner_gbfs_only_time does not work in multiprocessing batch processing. It gives 0 solutions with this, howver it worked with motion_planner_gbfs.
I have not used the one provide by jkljkljkl, but I suppose by changing the name temporarily from motion_planner_gbfs_only_time to motion_planner_gbfs should also work with his approach.
I did try all the permmutations with the names! Weird error!
Hi!
I just made a new multicore batch processing notebook. You could download it and run it locally.
I only changed the planner id in the config file, renamed the files, restarted the kernel - and it worked.
Tested with both motion planners, worked well.