DJI RoboMasters Mobile Manipulation Challenge

Organized by: DJI RoboMaster Committee
Website: TBC
Venue: Room TBC

DJI initiated RoboMaster in 2015 as an educational robotics competition for talented engineers and scientists. The annual competition requests teams who to build ground robots that use shooting mechanisms to battle with other robots, and the performance of the robots are monitored by a specially designed judging system, converting projectile hits into health point deduction of robots. The competition balances technology and entertainment. Many game elements in the competition are similar to video games. In August 2017, grand finals of RoboMaster 2017 attracted over 20 million viewers from over 20 countries around globe. Please visit​ to view game replays and introductory videos.​

In recent years, deep learning technology greatly has been reshaping the frontier of computer vision and many other AI research areas. In robotics research, deep reinforcement learning allows agents to make decisions basing on deep neural networks (DNN) instead of conventional hand-engineered finite state machines and if-else branches. Many excellent research platforms emerged from standard video games such as Doom, Dota and Starcraft. RoboMaster also has the potential to become a good deep learning platform, with robots that compete on the field fully commanded by DNN based policies. For one thing, robots also need to make hard decisions that cannot be done by conventional methods. For another, RoboMaster is a real game, so algorithms and policies developed on RoboMaster robots are more practical than those developed in virtual environments.

Illustration of robots need to be built (not the exact sample robot)

In ICRA 2018 DJI RoboMaster AI Challenge, team must build three RoboMaster Standard Robots and write autonomous programs for them (A sample robot will be sold by organizers, and a piece of code contains a skeleton of autonomous robot will be open sourced). Organizers will present three AI robots to challenge team robots. Team robots are challenged to defeat AI robots on a simple 8mx8m game field within 5 minutes. In each challenge, neither team robots nor AI robots can be controlled by human operators. Winner of the AI Challenge is the team who defeats AI robots in shortest time or using fewest projectiles. According to the final ranking, teams will get certificates, DJI products and a cash price as awards.