ML Agents training change velocity and gravity, WHY?

Hey, i have a question regarding ML Agents. I programmed a 3D Flappy Bird Version and the Bird has a RigidBody. This works very well when i dont train the agent (Bird).
As soon as i start the training (“mlagents-learn” in anaconda console) the gravity and velocity changes.
The bird moves with transform.Translate(transform.right * moveSpeed * Time.deltaTime) and jumps with
rig.AddForce(Vector3.up * jumpForce, ForceMode.Impulse);
In training this is approx. 15 times faster than running the game without training.
Has anybody an idea how this is possible?

By default, during training, ML-Agents runs Unity at about 10x regular speed to speed up the training process. You can disable this when running by adding --time-scale=1 to your cmd command when starting the trainer.

1 Like

Nice, it works, thanks!