whl-cal
whl-cal contains 8 collected skills from wheelos-tools, with repository-level occupation coverage and site-owned skill detail pages.
Skills in this repository
Design guide for calibration, registration, estimation, and other iterative algorithms that need a baseline-first plan. Use this when asked to design or redesign a sensor calibration or optimization pipeline, especially when the request mentions data filtering, baseline, metrics, visualization, papers, open-source references, or incremental algorithm improvement.
Iteration playbook for improving calibration and registration pipelines without losing baseline comparability. Use this when asked to optimize a calibration algorithm, propose next iterations, run ablations, or improve quality step by step rather than making a risky one-shot rewrite.
Validation and visualization guide for sensor calibration and registration pipelines. Use this when asked to verify a calibration result, define metrics, add diagnostics, judge run quality, or decide whether an algorithm is trustworthy beyond solver convergence.
Benchmark and ablation playbook for comparing calibration methods fairly. Use this when asked to compare baselines, evaluate a new candidate, design experiments, or decide whether an algorithm improvement is real.
Failure analysis guide for unstable, inconsistent, or misleading calibration results. Use this when a calibration run looks wrong, diverges, flips between solutions, or passes metrics without being trustworthy.
Release-gating guide for deciding whether a calibration method is production-ready, review-only, or still experimental. Use this when asked whether a calibration algorithm should replace the baseline, be exposed to users, or stay as a research path.
Guide for turning papers, blogs, and SOTA-style ideas into safe calibration upgrades. Use this when asked to apply a paper, assess a novel method, mine open-source projects for ideas, or translate research into a practical roadmap.
Dataset and capture-design guide for sensor calibration pipelines. Use this when asked what data to collect, how to recollect a failing calibration dataset, how to improve observability, or whether a solver problem is really a data problem.