Round 1 Winners of NeurIPS 2020 Challenge: Flatland (Multi-Agent Reinforcement Learning on Trains)

August 18, 2020

Our team, “MARMot-lab-NUS”, is currently holding first place in the reinforcement learning (RL) category of the NeurIPS 2020 competition “Flatland”, aimed at developping novel methods to solve the multi-vehicle routing problem (a close problem to multi-agent path finding). This competition is open to both conventional techniques (Operation Research [OR]) and RL, with a strong emphasis on learning-based techniques (because NeurIPS). It is sponsored by the swiss national railways (SBB/CFF) as well as the German ones (Deutsche Bahn), and is organized by Aicrowd. The competition is still very much open and active, and teams do not have to have participated to round 1 to join round 2, which is starting soon.