Classical Control for Ground Vehicles
(In the form of an email to a student)
Very soon after the end of ENGR 205, the controls class at Stanford I was a course assistant for, I met one of the students from the class randomly at In-N-Out. We were casually talking about the class, what he was planning on taking in the future at Stanford, etc. He mentioned that he worked in the automotive industry, working on software but closely with controls engineers. He took ENGR 205 to learn more about the controls, but instead he felt more disconnected from the subject matter, and his experience in the course made him think “Maybe controls is pointless and I don’t like it”.
This email is what followed:
Hey <Student>,
I remember you mentioned at In N Out yesterday that you were taking controls because that's what you were interested in and felt like it would relate to vehicle controls, but now you feel less sure if you like the subject. That stuck with me. As someone who has used this subject directly for vehicle engineering in industry, I felt bad that you had a negative takeaway from the course. So here is my attempt to convince you otherwise.
I will say, when I took a feedback controls for the first time in undergrad, it felt extremely disconnected from application. Why was I doing these random calculations to draw Bode plots or root loci, messing with these transfer functions, which I just wouldn't work with in real life? This is until I first did stability analysis at work, and realized that Bode plots/Nyquist plots are extremely useful tools to understand how dynamic systems behave, where they go unstable, and how they can be manipulated (that is where controls come in).
For vehicle dynamics, the best example I have is the content in Vehicle Handling Dynamics by Masato Abe, you can look at Chapter 3 here: Vehicle Handling Dynamics | ScienceDirect. You will notice that when it comes to vehicle handling, there is some transient response that is well modeled using what we have learnt in class (settling time/overshoot etc). And the response to different sinusoidal inputs is modeled by Bode plots, a great example being steering angle input to vehicle yaw rate output, attached below. Once you have these dynamics, you can tune your car to get the attributes that you desire. In sports cars, these are all very finely tuned for peak performance. In F1, these are modified race-to-race by changing the suspension and thus changing the loading on the front/rear tyres, as per track conditions and driver preferences (not your usual "active control", but still need concepts of stability!)
There are many other parts (especially in an EV) where I see controls being used, such as powertrain control or active suspension (spring-damper system control, like your HW!). When it comes to autonomy, companies like Waymo will probably do the path planning with some higher level optimization methods (or these days, end-to-end neural networks), but actually controlling the steering and braking is best done with traditional control.
Long rant, and you are free to still not like the subject :) but wanted to share this material that really helps motivate the topics for me.
He replied and thanked me for sending all the info. I am not sure if I convinced him, but I’m glad I tried. It also resulted in this relatively compact post for why classical controls is important for vehicle dynamics and control. Of course, there are many non-linear edge cases where more complicated controls would need to be used (see this amazing work of engineering art, and lot of other work into autonomous vehicles using RL for control), but this is the preliminary control that every car needs (and has used since WW2).
Anshuk