// challenge brief · hack a ton 2026
Mental Map
From a single RGB camera, with no LiDAR and no depth sensor, build a map of a space the robot explored, then plan a route across it.
- // Proposed by
- Monsson
- // Industry
- Computer vision / robotics (Unitree Go2 Edu)
- // Difficulty
- 🔴 Advanced
★ Own jury, own €1,000 prize. Pick any one of the three to compete.
// the problem
Most robots map the world with expensive depth sensors. Doing it from a single ordinary camera is far harder, but if you can, the robot becomes much cheaper and simpler. This is the high-ceiling challenge: ambitious, and the one most likely to wow judges if it converges.
// your mission
From a single RGB stream, build a representation of the explored space (a 2D/2.5D occupancy map or topological graph) and plan trajectories between arbitrary points on it.
// how it works
- 1Reconstruct: build a map from an exploration video.
- 2Plan: find a path from point A to point B on that map.
- 3Act: output a command sequence (
walk X m,rotate Y°). - 4Bonus: recover metric scale and detect loop closures.
// what we provide
- 3–4 indoor exploration clips (office, hallway, room with obstacles).
- Ground-truth trajectories for evaluation.
// suggested approach
- Monocular visual odometry / SLAM (e.g. ORB-SLAM-style pipelines) or a learning-based depth+pose approach.
- Path planning over the resulting map is the easy half (A*, RRT) once the map exists.
- Floor version: a topological graph of places and a planned path between two of them. Perfect metric scale is a stretch, not a requirement. Get something navigable first.
// how you'll be judged
- Planned-route error vs. ground truth.
- Robustness to rotations and repeated passes through the same area.
- Visual quality of the produced map.
- Metric-scale accuracy (if attempted).
// stretch goals
- Metric scale from camera intrinsics and motion.
- Loop-closure detection to correct drift.
// deliverables
- Git repo: code + README + run instructions.
- Demo: notebook or script on the provided clips.
- 5-minute pitch: architecture, decisions, trade-offs.