jjmata / carputer

Carputer

Recording pipline

0. activate the virtualenv: source /path/to/venv/bin/activate
1. run a script to drive and record training data: python main_car.py record -- this will let you have manual control over the car and save out recordings when you flip the switch

Run autonomously

0. activate the virtualenv: source /path/to/venv/bin/activate
1. run a script to let tensorflow drive: python main_car.py tf -- when you flip the switch, you will lose manual control and the car will attempt to drive on its own
2. for autonomous kill switch: pull throttle and turn the steering wheel
3. to revive autonomous mode, hit the channel 3 button (near the trigger)

Training pipline

0. convert images to np arrays: python NeuralNet/filemash.py /path/to/data
1. train a model: python NeuralNet/convnet02.py
2. use this model to drive the car (see above)

Analysis

• for training info, see debug.html -- reminder: < 7 is right, > 7 is left
• run analysis/make_video.py for debug videos
• use analysis/plot_vs_time.py to view telemetry data

Hardware TODOs

• get new camera with no exposure bug
• securely mount camera to car
• front bumper
• Roll cage

Software TODOs

• make images 320 x 240
• throttle linear remap (in branch split_softmax)