The subject line reads “self-driving car”. That is not a one-person project, even if you work full-time for many years. But then I read the PDF. This is a classic hobby-robot project. A line following floor bot. — MUCH easier.
The hardest problem self-driving cars have to solve is predicting the future actions of other road users and, of course, route planning. But the floor-bot does not need to deal with that.
The hardest part of this project will be training the neural network. You need tons of data taken in different conditions. It is easy but takes hours. Well, it can be easy so long as you don’t want to handle the case where the car loses the line and needs to find its way back to the track. One thing you can do to cut the training time by 2X, 3x, or 4x is to install 2, 3, or 4 cameras each with a slightly different position, tilt and aim point. (Or use data augmentation to simulate this) After training, you go back to just one camera, but mounting an array saves you a lot of driving time. And yes, for training, you do want to move the camera after a few laps. You don’t want to over-train; you want the model to generalize.
As for battery, I like the ones made for drones. You get the best power-to-weight that way. They came in all different sizes and voltages. But you need a specialized charger and to know a little abnout batteries. But performance and cost are good. If I were building a “real” outdoor robot that was going to be reliable and easier to use. I’d use DeWalt or some other kind of power tool battery because they are easy to charge and swap. Buy whatever brand you already use. You get the best quality and reliability by far
You are going to want decent battery life because you will need to drive this robot for hours to train the model.