SBB, T-Systems and ZHAW are establishing the Flatland Association.
Our research group is finally announcing its partion as research partner respresenting the ZHAW in the newly established Flatland Association. In collaboration with the SBB CFF FFS and T-Systems, the Flatland Association faces the resource challanges of contemporary railway networks with innovative solutions. The newly founded association sets the corner stone for collaboratively navigate the future challenges of transportation networks.
About Flatland
Flatland’s goal is to enable industry partners and academic researchers to achieve an interdisciplinary approach to work, in order to implement innovative solutions for the industries at hand. Flatland started with various challenges on AIcrowd in 2019, when several teams tried to solve the vehicle re-scheduling problem (VRSP) or, more generally, the re-scheduling problem (RSP) with multi-agent reinforcement learning. This is a problem faced by many transport or logistics companies around the world, including SBB, which in 2019 operated the most densely mixed railway network in the world. Like almost every company in the transport sector, SBB is facing ever-increasing demand and needs to cope with a large increase in capacity. This significant increase in capacity must be achieved either through denser train schedules, new infrastructure or investment in new rolling stock. One solution to assess the impact of different capacity increases is to simulate the railway infrastructure. SBB’s research group has developed a high-performance simulator that simulates both the dynamics of train traffic and the railway infrastructure.
Figure: Visualization of the Flatland simulation environment
What’s next?
Flatland’s mission is to foster innovation and open the way for new solutions to the resource allocation problem by applying the existing and continuously developed Flatland simulation environment. It is our vision that Flatland becomes a hub for open research, providing space for highly specialised, but also cross-disciplinary exchange. To this end, Flatland provides a rigorous technical problem formulation and a standard benchmark that is recognised by both the industry and the academic research community. For more visit the website of the Flatland Association or get further information via Flatland on AIcrowd.
Scientific Publications
- Flatland-RL : Multi-Agent Reinforcement Learning on Trains
- Flatland Competition 2020: MAPF and MARL for Efficient Train Coordination on a Grid World