NovelBot

The goal of this little project is to train a simple robot to navigate it's way out of a room (or maze in the future) using novelty search.

Right now, I have a simulation of a several robots that bounce around a room and that explode when they touch one of the vertical walls. The next steps are to:

• Give the little robots a brain
• Evolve the brains with a genetic algorithm and a fixed fitness function
• Evolve the brains using novelty search

All the code is done in Haskell at the moment using the FunGEn library for making games. This is most likely not the best or most efficient way to code this thing, but it's fun!

Simulation

Roadmap

• Make the walls
• Make the robot
• Make the robot explode when it touches the vertical walls
• Ensure that if the robot speed is large it cannot just jump over the walls

Braains

So how do we give the little robots a brain? There are many ways to model this (going back to the Explaining Intelligence course) but to make things simple we will stick to the Sense-Think-Act model where each time step the robot gets to:

1. Sense: the robot gets to look in 4 directions around it and gets the distance to the nearest wall
2. Think: processes the information from its senses
3. Acts: modifies its speed property

Roadmap for the little robot brains

• Make the robots gather accurate sensory data
• Implement noop sense -> think -> act loop
• Make a simple brain so that the robot no longer touches the walls
• Make the brain into a neural network and encode it in a genome
• Replicate the simple brain with the neural net

Sensing

In our case, the robot has 4 "eyes" pointing north, south, east and west. Each eye essentially measures the distance between the robot and the closest wall in that direction. The distance is either finite if there is a wall in that direction or infinite if not.

                                  N
                                  ^
                                  |
                            W <-- R --> E
                                  |
                                  v
                                  S

Thinking

To begin with, the brain of the little robot will be a simple linear function of the distance to closest walls and its speed. The output is new speed for the robot. Therefore, the brain will be represented by a total of 12 numbers (6 weights for each speed output).

Inputs: the 4 distances to the nearest wall in the cardinal directions (N,S,W,E) and the 2 speed coordinates (X,Y). 6 total inputs.

Outputs: the 2 new speed coordinates (X',Y').