Research Spotlight: Vision-Based Autonomous Navigation for Martian Terrain
A look inside the perception pipeline that lets our rover see, map, and traverse unstructured terrain without GPS — and the research behind it.
Autonomy is where competitions are won and lost. In this research highlight, our autonomous navigation team breaks down the vision-based pipeline that allows our rover to localise and plan paths across rocky, GNSS-denied terrain that mirrors the Martian surface.
From Pixels to Paths
The system fuses stereo depth, visual-inertial odometry, and semantic segmentation to build a local traversability map in real time. A sampling-based planner then computes safe trajectories around obstacles, while a recovery behaviour handles the inevitable edge cases when the rover loses confidence in its position.
This work underpins a forthcoming paper from the team and continues to evolve with every field test. We believe sharing it openly strengthens the wider student-robotics community.