Hi all,
I hope your code deals with cars stopped in the start line because I think that's what our car is going to do in test session 2 :p
We used some libraries in our code and we forgot to ask this before: What should we do to include some libraries? Can we request for them to be installed or should we include them as if they were part of our code?
Thanks,
RasPerras del Infierno
The standard code should try and drive around the obstacle, we will see if this works on race day :)
If you are including additional libraries not included in our standard SD card image then you have two options listed below.
As this is just a testing round we should be able to take a late submission from you as long as it is before Monday if you want to try and sort the libraries out.
If you do send us an email to info@formulapi.com to let us know we need to rebuild the SD card for your entry.
1. Include them in your code
If they can be placed in the
~/formulapidirectory with the rest of the Race Code then they can be uploaded as a part of your code.The easiest way to test this is to create a standard image following our Standard Formula Pi SD card image instructions.
Once you have done that copy your code into the
~/formulapidirectory and check if everything works.2. Send us a pre-setup SD card
If the libraries you are using need to be correctly setup or are very large it is probably better to setup your own SD card and mail it to us.
This will allow you to setup the SD card however you like :)
What we will do is overwrite the
~/formulapidirectory with your FTP upload and empty the~/formulapi/logsdirectory each round.This allows you to still make changes even if you send us a custom SD card.
Thanks! We have uploaded the new version which only adds the library as part of the code.
Let's hope it works :) See you later in the streaming!
RasPerras del Infierno
Okay, we have re-written the SD card with your new upload, we will see how everything goes later on :)
Sorry again!
Please, please, use the recently uploaded version if possible! We had done some more tests and the neural network in the previous file was wrong :( I've recovered the one from the original submission.
Sorry for the inconvenience and thanks again!
No problem, we have time to re-write the card again :)
I will let you know when it has been done.
:) Thanks!
Your latest FTP upload has now been written to the SD card.
Hi,
We are still fighting with including our libraries as part of the submission :-s
Would it be possible to send one "apt-get install a b c" command to run on our SD card?
Thanks,
Jorge
We could try making a custom SD card load for you here.
We cannot really make it part of the normal process as the SD card is imaged offline.
What was the apt-get line you need to run?
This would do for us. The rest of the dependencies can be included in the zip file.
sudo apt-get install python-scipyThanks!
Jorge
Okay, I will setup a custom SD card for you with python-scipy installed.
Have you uploaded a version we can test to see if it works correctly?
Hi,
We have uploaded a file just now. We really hope this one works if scipy is installed. We'll maybe update some update before the deadline, but the usage of the libraries shouldn't change much from now.
Thanks!
Jorge
The test file is not decompressing properly:
gzip: stdin: invalid compressed data--format violatedIs it possible the file was uploaded in ASCII mode instead of binary mode?
I uploaded again. It could be the binary setting, I forgot it before. Today filezilla decided not to connect to the ftp server, and we reviewed the file manually so we didn't use the fpi-upload script.
Sorry about that :(
Jorge
That version unpacked and ran successfully :)
Cannot wait to see how it does on race day.
Finally! Thanks so much!
Now it's time to optimize the neural network and so on. We have lots of work to do there too!
Cheers,
Jorge
Good luck Jorge... I'm really looking forward to seeing how your car performs with a neural net as the driving force: that should be awesome!
Geoff
If you saw the race: we have lots of work to do. We think the current version takes too long per image and that makes it turn too much... In the simulator and running in a laptop (faster than the pi zero) it worked fine though.
Let's see if it gets better in the next races!
Hi Jorge,
Yes I saw the race, and I think that most of us saw unexpected happenings as we gazed at those heats.
I might have done apparently well on my heat, but working through the logs and image captures that I got back I realised that my car had come to a premature halt because it thought that it had completed 23 laps! Extra laps occur particularly when a collision with another vehicle happens: anything that causes the camera to tilt down towards the track has the start line detection picking up the red track as a false start line. I did get enough information to know that some of the algorithms I was trying to use were definately bad choices, so there's be quite a few changes before the next round.
As for your neural nets, have you tried running some timing trials on a PiZero to check what sort of time you're getting? I have to admit that it did cross my mind to wonder how you were achieving a good turn around, because when I've modelled any nets they generally get pretty intensive in the calculations. I'm guessing that you're creating the actual net in Python and using the SciPy library functions for the number crunching.
If it needs speeding up, have you considered implementing the full net processing in Fortran or C?
Purely a suggestion made out of ignorance! :-)
Good luck to everyone!
Kind regards,
Geoff
Add new comment