The current implementation treats the entire gate as a point mass and the success condition is that the distance between the ball and the gate center is smaller than a threshold. This is unrealistic because normally one would assume that as long as the ball
1) passes the gate line, and
2) is between the two gate posts
Then it's a goal.
So this PR changes the success condition to the above. As I checked, no existing example or the readme will be affected by this PR. Note that the existing reward shaping that's based on the distance between gate center and ball can still be used as an approximation.
Compared to the existing setup: The existing setup has a default value of 0.5 for the success distance threshold. Since the gate width is 2.0 then there will be certain areas near the two posts that won't be counted as successes even if the ball actually rolls into the gate.
The current implementation treats the entire gate as a point mass and the success condition is that the distance between the ball and the gate center is smaller than a threshold. This is unrealistic because normally one would assume that as long as the ball
1) passes the gate line, and 2) is between the two gate posts
Then it's a goal.
So this PR changes the success condition to the above. As I checked, no existing example or the readme will be affected by this PR. Note that the existing reward shaping that's based on the distance between gate center and ball can still be used as an approximation.
Compared to the existing setup: The existing setup has a default value of 0.5 for the success distance threshold. Since the gate width is 2.0 then there will be certain areas near the two posts that won't be counted as successes even if the ball actually rolls into the gate.