How to create a motor controller for diff drive robot?

[Bob-The-Worker]

Hello everyone, I’m working on a Barakuda robot from Shark Robotics, which is a robot controllable in differential drive, with deformations at the wheel level and deformations due to the suspensions. These various deformations make rotational control complicated and highly unpredictable depending on ground friction. Using the standard control available with ROS, I can navigate the robot, but not with enough precision.

Since reinforcement learning is becoming more and more prominent, I started wondering if, for motor control with physics like my robot’s, it could be useful. I would like to send a linear command in x and an angular one, and have the network return the 4 motor commands required to follow the input command.

Do you think this is feasible? Is it possible to greatly improve performance?
Thanks for your feedback.