The core of the competition is to train a robust low-level walking policy in simulation, then perform sim-to-real transfer on an actual Unitree Go1 quadruped robot using the Isaac Lab framework.
Once that works, you'll have the opportunity to code a high-level, vision-based navigation controller to drive your robot autonomously in a simple obstacle race using monocular fisheye camera images. To make this task manageable, we will place AprilTags in the arena.
On Friday, the final challenge will evaluate your walking policy across three tiers of difficulty:
No obstacles, joystick control - Navigate a straight line to reach the goal as quickly as possible using joystick control.
Obstacles, joystick control - Same as Tier 1, but with obstacles like walls and uneven terrain in the arena.
Obstacles, vision-based control - Same as Tier 2, but without joystick control. Your robot navigates autonomously using images from a front-facing monocular fisheye camera. AprilTags with known positions are placed around the arena and near the goal, while tags with unknown positions mark large obstacles.
Good luck everyone!
Link to the challenge repo for more details: https://github.com/MontrealRoboticsSummerSchool/MRSS_2026-go1-challenge/tree/main