Self Driving Car Simulator for Udacity Challenge #2
This is to help evaluate your models. Given the model and images, steering angle, speed, time, the simulator will try to show how the car will move using the predicted steerings.
- OpenCV
- Scipy
- Numpy
- Pygame
Check out sim_runner.py for example usage
At instance t0, the heading for both real and predicted car is the same (0 degree)
We get real_steering and predicted_steering from this step
At t1, we calculate the heading caused by each of the above steering angles (using speed at t0 and t1-t0). The difference in these angles (theta_error) is the amount that we will rotate our image. That is, if the car was steered according to prediction instead of the real thing, the image visible to the camera will be rotated by this much.
At t2, we again calculate the above diff and add it to theta_error. This will be the new rotation
A failure scenario is reached when either
- Rotation of the image is not possible
- An upper limit of rotation angle is reached
We could also have used net horizontal displacement for crashing out (like in the NVIDIA paper).
Also, I am not using horizontal displacement in the simulator as of now. May be in the next version I will use the left and right images and stich them together to get translation error too.
Reach out to me (@yousuf) on Udacity's Slack Channel
Adapted from comma.ai's view_steering_model.py