Subproject leader:  Mel Chua

Playing martial arts video games is fun. Playing martial arts video games for 2 fights for each of 1024 robots for 99 generations is tedious. In order to evolve the smartest robots possible, we needed to let them think for themselves and fight automatically. 

The default AI turns the default gene (a 45x45x44 matrix of integers between 0 and 100 inclusive) into a nondeterministic finite state machine. During each timestep, the robot determines the state it is in and the state its opponent is in; for each combination of those two states, the gene has an array of integers, each representing the probability that the robot will transition to a certain state next. The robot randomly chooses from among these options, executes its choice, and the program chugs on.

There are plans to create new versions of the gene and AI, including a neural network, a parameter-based intelligence, and a learning robot (that adapts during the course of a fight).