This lab works on phones, but writing code is much easier on a laptop.
Localization & SLAM · LiDAR SLAM · M5-L5 · Advanced
The overnight mapping run produced a map with a doubled hallway. All you have are the logs. The residual looks fine, which is exactly what makes it hard.
You find which of three faults happened and where it began, then decide how the pipeline should have refused to accept it.
You have passed every stage. Now connect it back to the theory and move on:
Plain Python with math and, for matrices, slamkit (matmul, transpose, solve, inverse, identity). The data arrive as the dictionary D. There is no NumPy, random, json or copy in the browser runtime. Your code runs locally and never leaves this page.
Stage 1 is a diagnostic: each fault of M5-T7 and M5-T10 leaves a signature in a different column, and the grader checks the cause and the first symptom on hidden runs with different faults and decoy events. Stage 2 is the corresponding prevention: a fit gate and a consistency test for loop candidates, tuned on a set that includes a look-alike place.
A mapping system that cannot say why it failed cannot be improved. The skill of choosing the metric that separates the hypotheses carries over to every other SLAM failure.
Educational simplification: the logs are per-keyframe summaries generated from a fault model with exactly one fault per run, and the loop candidates are described by their ICP fit and a consistency flag. Real failures overlap, metrics are noisier and ground truth is rarely available.
Get notified when new robotics labs are released. One email per new lab, nothing else.