Feedback and Suggestions

[PuzeLiu]

Hi everyone,

We would like to hear your feedback about the sessions. You can either leave your comments and suggestions here or send us separate emails! Thanks!

[AndrejOrsula]

Thank you all for your effort in organizing this challenge. It was an excellent opportunity to explore and experiment with new approaches in an engaging way. The challenge was well-organized, and I enjoyed the experience!

• The process of submitting Docker images was very streamlined from the perspective of a participant and made it simple to submit new solutions. I would love to hear your point of view on this matter.
• I believe the absence of a Gym/Gymnasium environment wrapper might have introduced a potential entry barrier for some participants and thus reduced the number of participant teams in the finals. It was not complex to write a custom wrapper - but at least my implementation is very ugly/hacky and contains unnecessary duplicates for training and evaluation. Having an official wrapper for a popular API could have simplified participation.
• I believe there is a bug in the tournament environment reset in cases where the puck gets stuck in the middle. As per the rules, the puck should reset on a random side of the table. However, it currently resets on the same side that the puck started last time. This may result in only one of the participants ever being able to interact with the puck. (I discovered this after training an agent that received only a negative reward for receiving a goal and causing a foul. However, the agent learned to optimize a behaviour that always made the puck get stuck on every reset. This way, its opponent would never touch the puck, and the risk of receiving a goal would be minimal.)
  • Adding self.start_side = np.random.choice([1, -1]) somewhere around this line (example) would fix this issue in cases when the puck gets stuck and also when the simulation explodes (when the puck is too fast or outside of the table boundaries).
• With respect to the simulated "loss of tracking" during tournament evaluation, there could potentially be cases where a controller would never observe the puck moving towards their goal if the puck was shot at a very high velocity at the beginning of the loss of tracking. I understand the reason behind including this aspect of the environment, but it might make the evaluation more stochastic.
• There is a ruleset referenced in the description of tournament. However, I believe these rules are outdated and do not fully correspond to the code and other descriptions of the rules (win condition, duration until foul, and area where the puck can get stuck).

Thank you again for organizing this challenge!