roverslip op• Data kinds: signal × signal → table
• Call: import fullseye as fs; fs.ledger.slip_quantile_fit(slopes, slips, quantiles=None, n_knots: 'int' = 6, smooth: 'float' = 0.001, iters: 'int' = 200, calibrate: 'float' = 0.25, level: 'float' = 0.9, seed: 'int' = 0) -> 'dict' (to call the implementation directly, import roverslip; roverslip.slip_quantile_fit(slopes, slips, quantiles=None, n_knots: 'int' = 6, smooth: 'float' = 0.001, iters: 'int' = 200, calibrate: 'float' = 0.25, level: 'float' = 0.9, seed: 'int' = 0) -> 'dict'; from the registry, opsdrive.get("slip_quantile_fit"))
Quantile regression from slope angle to slip ratio (piecewise-linear basis, pinball loss via IRLS, weak penalty on 2nd differences).
The default quantiles is 19 levels 0.05, 0.10, …, 0.95. n_knots knots are placed at quantiles of the angle (so as not to place too many knots at sparse ends of the data). After fitting, the quantile curves are sorted per angle to remove crossings (rearrangement). Even when the noise varies with angle, the band width at that angle follows it — the difference from the Gaussian process's uniform noise. If calibrate > 0, that fraction of the data is set aside for calibration (a random split determined by seed), the rest is used for fitting, and the width of the two-sided level band is corrected by conformal prediction (CQR of Romano, Patterson & Candès 2019: for each calibration point take E = max(下端 − y, y − 上端) (下端 = lower end, 上端 = upper end), and widen / narrow the band on both sides by its ⌈(n + 1) level⌉ / n quantile). For exchangeable data the band coverage is at least level in finite samples (a guarantee of overall coverage, not per-angle coverage). IRLS quantile regression tends to estimate the band too narrow in-sample (82–91 % for a nominal 90 % in the prototype measurements), so it is enabled by default. Returns {"kind": "quantile", "quantiles", "knots", "coef": (段, 節点), "conformal": {level: 幅の補正}}.
• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).
• Operator provenance and references — the sources of the research/methods this op family came from.
• The canonical algorithm (author, year) and its uses are named in the family usage guide above.
• poc_rover_slip_risk_path — py -3.11 examples/poc_rover_slip_risk_path.py
table as input)course_layout · course_occupancy · course_contains · world_build · world_camera · world_move · lidar_scan · rss_longitudinal_same
roverslip)bekker_pressure · bekker_wheel_sinkage · wheel_forces · wheel_sinkage · wheel_traction_curve · slope_slip_curve · ground_shift_track · odometry_slip
*Provenance: roverslip.py — DRIVE operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.