Hi - I’ve been trying to use MLagents to solve this discrete hide and seek game but with limited luck. I would really like to get some suggestions on what I could improve.
The premise of the game is that the red agent hides in one of 3 locations defined by each of these corridors. The red agent leaves breadcrumbs that the blue agent can perceive. The blue agent is supposed to spot the red agent to win the game. The input that the red agent gets is its position and knowledge of the surrounding breadcrumbs in the 8 surrounding tiles. Once the seeker sees a breadcrumb it goes away. I give the seeker 1 reward for finding a breadcrumb and 100 for finding the hider. Specifically what the seeker fails to do is use the breadcrumbs to find the hider. If the hider takes the same route every time the seeker can find it.
