Benchmark Visual SLAM Stacks for an Indoor Delivery Robot
Overview
What this challenge is about.
You receive 8 indoor rosbag recordings (about 90 minutes total) captured by the robot's stereo camera + Inertial Measurement Unit (IMU) plus ground-truth trajectories from an external motion-capture setup. Run ORB-SLAM3, OpenVSLAM, and a learning-augmented baseline (e.g., DROID-SLAM) on all 8 bags, measure Absolute Trajectory Error (ATE) and per-segment drift, and characterize failure modes (re-localization after kidnap, long corridors, dynamic obstacles). Deliver a benchmark report with one clear recommendation backed by per-scenario numbers.
The Brief
What you'll do, and what you'll demonstrate.
Pick the visual-SLAM stack with the best accuracy-vs-robustness trade-off for indoor hospital corridors and defend the choice with reproducible numbers.
Earning criteria — what you'll demonstrate
- Run and evaluate modern visual-SLAM systems on real robot data
- Quantify localization quality with ATE and segment-drift metrics
- Characterize failure modes in long-corridor and dynamic environments
- Defend a perception-stack choice in writing to engineering leadership
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
Benchmarking visual-SLAM systems on real robot data with rigorous metrics is a core junior CV-engineer task at any indoor-robotics company; this challenge gives the student a defensible portfolio project.
This challenge sharpens
- visual-slam
- sensor-fusion
- benchmarking
ML Researcher
Designing a fair comparison across classical and learning-augmented SLAM exercises the same controlled-experiment muscle used in applied-research roles.
This challenge sharpens
- visual-slam
- benchmarking
- trajectory-evaluation
AI Engineer
Wrapping three research-grade SLAM stacks into one reproducible harness mirrors the integration work AI engineers ship at robotics startups.
This challenge sharpens
- python
- ros
- benchmarking