Hi guys, i am new to UnityML, so currently i am working on a project which is like ‘Pursuit Evaders’. So the pursuer needs to find the evaders in the environment as quick as possible. However, i want to declare the observation that at the start there is no evaders in the their line of sight. Only when the evaders is their line of sight, then we include the evader’s distance/bearing etc. Anyone got any ideas how to do it?
The easiest solution would be to make your observation space as big as it might be needed. All the values that are not used could be simply filled with zeros or some other constant value that suits your observation space best.
I did tried with some constant value, whereby everytime once it reset the game, it will be set as 0.5 for distance, bearing/orientation to be 0.5 * pi. If the agent didnt sense the evader, the observation will also be set as 0.5 for distance, bearing/orientation to be 0.5 * pi. I realize that the training time is longer, and it converge slower, is it normal for these to happen? thank you so much for your reply
Are you using the RayPerceptionSensor?