A blind Pacman hunts ghosts with a particle filter and a greedy search algorithm. At each update step, Pacman "listens" for the ghosts and receives noisy data of their locations. Pacman then distributes particles throughout the maze and moves towards where he thinks the closest ghost is. The brighter the square is, the more particles there are on that square and the higher the probability the ghost is there. If Pacman runs out of particles for any ghost, he starts over and resets all of the particles. I worked on this with Yue Chang Hu as part of a project for the UC Berkeley class "CS188: Artificial Intelligence".

You can see more particle filter trial videos on my YouTube chanel: 1 2 3 4 5