My current personal project is to build and implement Artificial Intelligence into a Robot Arm. I’m using it as a test bed to try out computer vision algorithms with reinforcement learning in a hardware product.
The inspiration for this project came from observing robots attempting to grasp objects. That problem has a massive number of factors and design considerations, and I wanted to understand better a more basic one: teaching a robot arm to move an object in a linear direction.
I am setting up the experiment like this: The arm has 3 axes of rotational freedom, and one clamper. I am setting up a rod 5 inches away from the base of the arm, and the rod is attached to a flex sensor. As the rod is pulled up, the flex sensor sends an analog value that serves as a heuristic for success to the computer. There is also a camera over the whole setup.
The genetic algorithm will encourage the arm to get close to the rod, and pick it up. The inputs it receives are the overhead video feed, and the distance the rod is moved up. I will update this post as progress continues.
Huge thanks to daGHIZmo for the original designs for the robot arm.