Your submission is now in Draft mode.

Once it's ready, please submit your draft for review by our team of Community Moderators. Thank you!

Submit Essay

Once you submit your essay, you can no longer edit it.

Pending

This content now needs to be approved by community moderators.

Submitted

This essay was submitted and is waiting for review.

What will be the score of the best performing agent in the Flatland AI competition?

Question

Can an RL system get the trains to run on time? The Flatland Challenge is a competition in multi-agent RL tackling the Vehicle Re-Scheduling Problem (VRSP), the problem of how to efficiently schedule transit routes while taking into account real world conditions like breakdowns and delays.

Sponsored by the Swiss Federal Railways - who manage the densest mixed railway network in the world - the goal is to create RL agents that can schedule trains in a gridworld environment which minimize the total time of all operating trains.

The first round was focused on basic scheduling - how to get multiple agents (trains) to their destination quickly. The current round has a more complex problem set, with trains: - Operating at four different speeds - Entering and leaving the environment - Randomly malfunctioning, rendering them immobile for a period of time.

The best systems will be able to plan ahead and route efficiently around malfunctions.

As of Oct 28th the best performing agent on the round two leaderboard had a normalized reward of -23.010.

What will be the normalized reward of the top performing agent be at the end of the competition?

Resolution Criteria: The outcome will be determined by the leaderboard at the end of the competition.

The normalized reward is a function of: - Mean number of agents done, in other words how many agents reached their target in time. - Mean reward is just the mean of the cumulated reward. - If multiple participants have the same number of done agents they compute a “nomralized” reward as followes: normalized_reward = cumulative_reward / (self.env._max_episode_steps +self.env.get_num_agents()

Make a Prediction

Prediction

Note: this question resolved before its original close time. All of your predictions came after the resolution, so you did not gain (or lose) any points for it.

Note: this question resolved before its original close time. You earned points up until the question resolution, but not afterwards.

Current points depend on your prediction, the community's prediction, and the result. Your total earned points are averaged over the lifetime of the question, so predict early to get as many points as possible! See the FAQ.