This lab works on phones, but writing code is much easier on a laptop.
Localization & SLAM · Advanced SLAM and Research · M8-L4 · Advanced
The vendor’s slide says “0.12 m average error”. Your own tests show that one run in eight ends in a map that is unusable. Both statements are true of the same data.
You write the summary that makes the failure visible, then attack a real back end with false loop closures to see which defence deserves to be called robust.
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 turns the reporting rules of M8-T8 into a function: median, percentile and failure rate per condition and an identification of the condition to which the system is most sensitive, graded against the reference on hidden sets of runs. Stage 2 runs the pose-graph optimiser of Module 4 under injected false loop closures with and without robust kernels, twelve runs per setting.
Systems are chosen on summary numbers. Whether those numbers include the failures decides whether the choice is sound.
Educational simplification: the system under test is the SE(2) pose-graph back end on a circular trajectory with 24 poses, not a complete SLAM front end; the failures are injected in the data (drift, false loop closures, no loops); twelve runs per condition, a fixed failure threshold on the ATE. Real studies also inject sensor dropouts and timing faults, use many more runs and report confidence intervals.
Get notified when new robotics labs are released. One email per new lab, nothing else.