chaupow / carnd-pid-control-project

CarND-Controls-PID

Self-Driving Car Engineer Nanodegree Program

---

P, I and D

This section explains the effect of the P, I and D parameter when used to set the steering angle of a car that should follow a defined trajectory:

• The P component in the PID controller sets the steering angle towards the desired position proportionally to the error. This means, the bigger the parameter, the more aggressive will the controller head to the goal.

• The D component in the PID controller sets the steering angle stands for derivative. Depending on how big the change (therefore derivative) of the error is, the steering is being damped/strengthened.

• The I component (integral) increases steering if the error persists over time. This is necessary if the steering is slightly skewed.

Parameter Tuning

I started to manually tune P, then D then I, resulting in 0.03, 0.00001, 6 as PID values. These parameters were able to move the car along the track for a whole round.

I then implemented the twiddle algorithm that was presented in the class. The challenging part was to simulate the control flow that was done in the class with run(robot, p) within the simulator.

I first ran twiddle with [1,1,1] similar to the class but noticed that (as I started with parameters that seem to be ok) 1 was a bit much and decreased dp to [0.1, 0.1, 0.1].

The final values are [0.577838, 1e-05, 7.48411]. Watching the car drive with this PID controller, it feels a bit shaky. I assume that the reason is the high D value. I think this could potentially be fixed by incorporating a very high steering value into the error but I haven't tried this out.