ERF Robotics Hackathon: Autonomous Robot Navigation

F. Barkhof, S. van den Bent, O. Bosgraaf, S. Gunneweg, J. de Heus, T. Nijdam, D. Prinzhorn, J. Sprott

ERF Robotics Hackathon barn arena with Lely Juno robots

Summary

At the European Robotics Forum 2022 "Battle of Institutions" Hackathon in Rotterdam, our UvA team programmed two Lely Juno robots to autonomously navigate a simulated barn environment. The Lely Juno is an automatic feeding robot designed to push back hay to cows in real barns.

The challenge consisted of 9 scored tasks including wall following, narrow passage coordination between two robots, obstacle avoidance, homing to a base station, and proximity-based light activation. We competed against 7 other European university teams on challenges from Lely, Franka Emika, and Robo House.

Implementation

Dual navigation strategies. One Juno used a behaviour-based wall-following algorithm with three sonar distance sensors, switching actions based on right-wall and front distances. The other used a model-based approach with a LiDAR scanner for SLAM (Simultaneous Localization and Mapping), creating a cognitive map and following waypoints via a finite state machine.
Multi-robot communication. A ROS master-client architecture where robots coordinated passage through narrow corridors using boolean messages. The master Juno had priority; the client would stop and back up when the master was in the passage.
Color-based homing. A blob detection algorithm using an external webcam to locate the home base (an orange traffic cone). The center of the detected blob determined left/right steering adjustments. At the ERF, the detection color was switched to green to avoid confusion with the many red objects in Lely's arena.
Vicinity lights. Blob detection was reused to estimate inter-robot distance. When one Juno detected the other's color within range, IoT-connected Shelly Plug-S smart plugs were triggered via MQTT to turn on lights.
Developer monitoring. ROSboard provided a real-time web dashboard showing LiDAR scans, costmaps, sensor readings, and robot state.

Challenges

The LiDAR could not be center-mounted on the Juno due to its physical design, limiting the scan to less than 360 degrees and degrading the SLAM map quality. This forced a switch from model-based to behaviour-based navigation for one robot. The second Juno lacked sonar sensor mounts entirely, so it navigated purely via camera-based blob following. These hardware constraints required significant adaptation of our pre-tested TurtleBot3 solutions during the hackathon itself.