koobas.hobune.stream
Discrete Rewards or Continuous Rewards? ML-Agents PPO
Which of the two forms of reward is better and why?
1. AddReward(Mathf.Cos(_rigidbody2D.rotation * Mathf.Deg2Rad));
2. if (Mathf.Cos(_rigidbody2D.rotation * Mathf.Deg2Rad) >= Mathf.Cos(10 * Mathf.Deg2Rad)) AddReward(1);
I perceived that the first form converges more faster.
