AutoRCCar icon indicating copy to clipboard operation
AutoRCCar copied to clipboard

TURN RIGHT

Open hobnob453 opened this issue 5 years ago • 18 comments

I have followed the instructions as part of a college project, after managing to set up all of the python code, i have trained the car with 6 .npz files which modify the nm_model.xml, however when i run the main program (rc_driver) along with stream and ultrasonic files on the PI, the car just wants to turn right all of the time, i will continue to add training data but could do with some guidance as to where i could be going wrong? Thanks,

hobnob453 avatar Mar 05 '19 18:03 hobnob453

i also face the same problem.......

kxrf97 avatar Mar 06 '19 16:03 kxrf97

did you manage to tackle it?>

kxrf97 avatar Mar 06 '19 16:03 kxrf97

Having another go tonight, just gonna train it some more, glad its not just me doing something stupid, how much ‘collect_training’ have you done?

On 6 Mar 2019, at 16:22, kxrf97 [email protected] wrote:

did you manage to tackle it?>

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

hobnob453 avatar Mar 06 '19 18:03 hobnob453

I think something wrong with your wheel alignment maybe and check in the Python shell for forward commands and inspite of forward commands sent your car turns right then it's wheel alignment problem!

sundar19 avatar Mar 07 '19 03:03 sundar19

I collect around 300 images, and my accuracy is around 60+ after training.... But my car just keep going randomly

kxrf97 avatar Mar 07 '19 06:03 kxrf97

@sundar19 Wheel alignment are having no problem, the command prompt keep showing right , should not be the alignment problem. Did you manage to finish this project.

kxrf97 avatar Mar 07 '19 06:03 kxrf97

Yes my shell is sending right commands, only right commands

Sent from my iPhone X

On 7 Mar 2019, at 06:23, kxrf97 [email protected] wrote:

@sundar19 Wheel alignment are having no problem, the command prompt keep showing right , should not be the alignment problem. Did you manage to finish this project.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

hobnob453 avatar Mar 07 '19 06:03 hobnob453

@kxrf97 I used Cascade classifier of left direction sign to finish my project . The neural network you use will be efficient only when the streaming rate is good. Switch off the car and move the car by hand whether the camera lags,if that's the case Cascade classifier is the only solution

sundar19 avatar Mar 08 '19 05:03 sundar19

Can you send me any tutorial about that? How you use cascade classifier to finish the project. My streaming rate is abit laggy... I guess no smooth as i think...

kxrf97 avatar Mar 08 '19 05:03 kxrf97

Normally cascade classifier is use for object detection. How can you apply on the track

kxrf97 avatar Mar 08 '19 05:03 kxrf97

Yes i would love to know how you changed the classifier, again its not the smoothest but the connection should be fast enough.

Sent from my iPhone X

On 8 Mar 2019, at 05:58, kxrf97 [email protected] wrote:

Normally cascade classifier is use for object detection. How can you apply om the track

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

hobnob453 avatar Mar 08 '19 08:03 hobnob453

https://github.com/ryangmolina/rpi-car-arrow-detection-using-cascade-classifier?files=1

Take Cascade classifier for left or right sign from this link and include your right or left turn functions in rc_driver_helper.py

sundar19 avatar Mar 08 '19 15:03 sundar19

Yes I am facing the same issue, it always wants to make a 'right' turn. However on a different training set, it always made a 'forward' prediction. And on yet another one, where I trained it to only go 'forward' or 'left', it did output 'forward' and 'left' but the predictions were not correct at all. Not sure if its a training problem or something else.

somalwar avatar Mar 17 '19 05:03 somalwar

@somalwar have u solved that problem

VishalWrench avatar Mar 29 '19 15:03 VishalWrench

@VishalWrench No I haven't. I plan to use Tensorflow in the future, maybe that will give better results.

somalwar avatar Mar 29 '19 16:03 somalwar

@somalwar thanks for replying yeah i am also trying to implement this in keras

VishalWrench avatar Mar 29 '19 16:03 VishalWrench

@somalwar @VishalWrench Did either of you guys end up figuring out how to get the car to stop just going right all the time? was it an issue with the training?

jhadkaterji avatar Apr 01 '19 19:04 jhadkaterji

@somalwar @VishalWrench Did either of you guys end up figuring out how to get the car to stop just going right all the time? was it an issue with the training?

Im having the same issue and cant figure out whats wrong

jhadkaterji avatar Apr 01 '19 19:04 jhadkaterji