The experimental procedure is now quite similar to the experiment described in section 3.3.3.1. The difference is that, instead of using the joystick to teach the robot how to steer, the algorithm described above was used.
So in the learning stage, the robot stops for each frame. Then the corridor following algorithm which does not implement any learning is used to find a good steering angle. The Pattern Associator used in section 3.3.3.1 is then trained with these inputs and steering angle. The robot turns using this computed steering angle and then moves forward for a while before stopping in order to process the next frame.
No human intervention is required during the training. This can be very useful if the training stage takes a long time. Here, only a few tens of metre were required to teach the Pattern Associator. This teaching was done in several times because the robot was attached to its host computer with a wire that was shorter than seven metres. With a longer wire, we could have done the training in one step.
The testing stage is the same as it was in section 3.3.3.1.