Challenge

Bon Appétit! Feed the Robot Dog

The MRSS 2025 hands-on sessions and competition will give students the opportunity to apply the concepts and techniques learned during the summer school through a challenge involving multiple robots: a Go1 Quadruped Robot (Fido, the hungry dog) and a LoCoBot Mobile Manipulator (Alfred, the helpful robot butler). To successfully complete the challenge, teams of up to four students will need to work together to design, implement, and train policies that enable both of these robots to complete distinct tasks: the LoCoBot must collect and deliver food to the Go1’s dog bowl, and the Go1 must walk to that dog bowl to eat. The teams may choose the methodologies that they use to complete each task. However, any points earned by completing different tasks using the same cross-morphology policy will earn extra points.

Tasks

The competition will be divided into a series of distinct tasks for each robotic platform, each of which awards a different score upon task success.

Alfred’s tasks (LoCoBot):

• (1pt) Pick and Place – Perform fixed-position manipulation to pick up the dog’s food (a plastic croissant), and place it into a bowl.

• (1pt) Autonomous Navigation – Navigate the Locobot autonomously to a goal location.

• (2pt) Navigation + Pick and Place – Combine navigation and manipulation to pick up the food at a starting location, and drop it in a bowl at a goal location.

Fido’s tasks (Quadruped):

• (1pt) Walk Forward – Get your Go1 to take steps in a straight line.

• (2pt) Navigate to the dog bowl – Control the Go1 via teleoperation to walk all the way to a target location.

Scoring

Leading up to the competition, teams will have the chance to develop and test different methodologies/policies across all tasks on both robots.

During the competition, each team will have three chances (runs) to earn points, with each run consisting of both a quadruped task attempt and a LoCoBot task attempt. Points will be awarded based on task success, with the highest score from any run counting toward the final tally. 

The golden snitch: Finally, any run that uses the same cross-morphology policy to successfully complete both a LoCoBot task and a quadruped task will earn a 3x multiplier on the run’s total score.

An example team scoring card is as follows:

Starting software resources

All teams will have access to starting code repositories for:

• Isaac lab quadruped simulation for data collection, RL, and testing.

• LoCoBot code to deploy navigation policies onto hardware.

• CrossFormer starter code to train transformer-based cross-morphology policies.

Testing policies on hardware

Due to limited shared hardware, teams should develop their policies primarily in simulation, and book time slots for real-world robot testing as needed.

Thanks to InDro Robotics for generously providing Unitree Go1 robots for the summer school!