when using a rayperception sensor(2d), if i add additonal observations, the ray perception sensor no longer works.
this is the code i was using
// public override void CollectObservations(VectorSensor sensor)
// {
// sensor.AddObservation(health);
// sensor.AddObservation(totalWood);
// }
when using observations, the agent repeatedly runs into a wall, which it gets -reward for doing.
when not using these observations, it wiggles around normally.
is there a right way to do this?
also, when training 2 agents against eachother (different behaviors, i.e. hide and seek) is it best to use self play here? what about poca vs ppo?