Overview
What this challenge is about.
Use the public KITTI dataset (or a similar paired LiDAR+RGB dataset) restricted to static-obstacle classes. Implement a late-fusion baseline: a LiDAR-only detector (PointPillars) + an RGB-only detector (a small YOLO), then a fusion module that combines their outputs. Compare against an early-fusion variant (concatenated voxel + image features into a single backbone). Evaluate detection mAP + per-class breakdown + inference latency. Deliver a 3-page recommendation memo.
The Brief
What you'll do, and what you'll demonstrate.
Recommend between early- and late-fusion sensor architectures for indoor static-obstacle detection on accuracy, latency, and maintainability.
Earning criteria — what you'll demonstrate
- Implement two distinct sensor-fusion architectures end-to-end
- Evaluate detection performance with per-class and latency metrics
- Diagnose fusion-specific failure modes
- Recommend a perception architecture with trade-offs spelled out
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Computer Vision Engineer
Sensor-fusion architecture bake-offs are core CV-engineer work at every AGV, drone, and AV company.
This challenge sharpens
- sensor-fusion
- 3d-object-detection
- perception
Machine Learning Engineer
Disciplined comparison with per-class metrics + latency reporting is the MLE habit production teams expect.
This challenge sharpens
- pytorch
- benchmarking
- ml-pipelines
Applied AI Scientist
Translating fusion comparison into a maintainability-aware recommendation is the applied AI scientist's daily output.
This challenge sharpens
- sensor-fusion
- benchmarking
- perception