November 20, 2018 at 3:11 pm #30178
Long story short
Hi, previously I asked the question regarding leaderboard simulation – https://www.duckietown.org/forums/topic/leaderboard-time-paradox.
You can run experiment by yourself — compare 2 submissions ion the same straight line.
Scroll 1st place submission – https://duckietown-ai-driving-olympics-1.s3.amazonaws.com/v3/frankfurt/by-value/sha256/49629a6699f07585920a8417b8138b9b251777292473b33b711c0559ea877540 – from 13.3 to 15.0 seconds
Scroll my submission – https://duckietown-ai-driving-olympics-1.s3.amazonaws.com/v3/frankfurt/by-value/sha256/49629a6699f07585920a8417b8138b9b251777292473b33b711c0559ea877540 – from 7.7 to 10.8 seconds
I’m sure that at my submission the bot goes velocity=1 and steering=0. But it takes 3.1 seconds to ride the same distance, as 1st place submission, which takes only 1.7 seconds.
Turned out the wheel speed is not clipped!
So we took submission (1297) from 6 place with max wheel speed = [1, 1] and simply multiplied the wheel speed by 2 (submission 1304) — and that moved us to 2d place.
We kindly ask to fix this issue and clip the wheel speed, because for now it seems like all the solutions above 8 median tiles have the max wheel speed more than 1.November 20, 2018 at 5:04 pm #30185
Thank you for the note. We had in fact recently also discovered this problem. We are in the process of fixing. The speed will be capped to the equivalent of 2 tiles/sec.
Thanks.November 20, 2018 at 5:12 pm #30186
Liam, thank you for response. Could you elaborate on how 2 tiles/sec are related to DEFAULT_ROBOT_SPEED parameter in the simulator. Right now it’s 0.4.
Seems like the simplest solution will be to clip wheel speed in https://github.com/duckietown/gym-duckietown/blob/47d94df05417ef9a64e779e43202245c081f50b7/gym_duckietown/simulator.py#L1117.November 21, 2018 at 2:57 pm #30285
Here’s my proposal: https://github.com/duckietown/gym-duckietown/pull/118
You must be logged in to reply to this topic.