The computing division has started development on a new neural network, which will improve the robot’s autonomous performance.
“The neural network is basically like a graph of nodes,” computing division lead Chris Eitutis explains. “You feed it some input, and those nodes do calculations similar to what you would imagine your brain to be like… then it will give some output, which should be what we want the robot to do.”
The programmers will be using an open source library called TensorFlow to set up this neural network within the existing programming framework of the robot. Compared to traditional autonomous code, the new system will make it easier to prepare the robot for completely new locations, team member Shae Bolt said.
“It allows [the robot] to work with arbitrary math data that we don’t have to code specifically for, so it should theoretically… be more adaptive and better at navigating unknown courses.”
Once the neural network has been set up, improving the robot’s autonomous skills will just be a matter of training it with various maps to give it practice.