Introduction

This page contains links to gif animations of robots navigating hallways using evolved neurocontrollers. The controllers come from research done for my term project in Dr. Kuiper's Robot Learning class.

There are 4 movies included. The animation may progress quickly after pressing on the desired movie, so it is easy to miss the first few seconds of the animation if you don't know what to look for. In the first 2 animations, the networks start out at angles that are very far from the midline of the hallway, so they have to straighten themselves out. Notice they do this very quickly in both cases. Starting out from different angles makes the task challenging, since robots cannot just go straight and then stop when they see a wall.

The third clip shows the same controller from the first 2 clips. However, in the third clip, there is a door at the end of the hallway. Remarkably, the controller knows how to enter the door even though it was not trained for this purpose.

The final clip depicts a controller evolved to stop in front of the door instead of enter it.

Understanding the Movies

The movies are split into a left and right frame. The left frame depicts the robot in the hallway. The diameter of the robot is only as wide as the yellow circle (happy face) in the middle of the rings. A red line points in the direction of the robot, and the various sensor activation levels are depicted as red rectangles around the robot. The sensors rotate with the robot, so you can see what the robot sees in real time.

The right frame displays the neurocontroller for the robot, which was evolved using NEAT. The inputs, which come from the 8 sensors and a bias, are at the bottom of the network. The outputs are at the top. Neuron activation levels (including inputs and outputs) are depicted by red rectangles that expand and contract. You can watch the neural network "thinking" in real-time as the robot navigates.