The practical part of the Artificial Intelligence course at KTH (DD2380) on the fall 2021 was centred around the fishing derby game. TIt was completed three projects regarding the decision making of the AI system.
The goal of this assignment was to compute a few movements ahead using the minimax algorithm, assuming that the computer was always making the best move. In order so, the group used a heuristic function to approximate the theoretical utility function since the possible number of plays each round is too big. As we can see on the example, each fish has a different positive or negative value, and the boats cannot overlap each-other.
In this assignment, we had to predict the type of each based on their pattern. The only thing known was that there were 7 fish speacies, although each specie might has multiple swiming patterns. To correctly identify the speacies, we ahd to formulate a classifier in real "swimming" time.
The reinforcement learning assigment concerned a diver swimming until the king fish, avoiding the the jellyfish. We used the Q-Learning algorithm with an epsilon-greedy policy. The agent was rewarded +20 when he reached the king fish, -10 by touching the jellyfish, and -1 per step.