Repository navigation
Hydraulics and bucket calibrated against PWRI's ZX200 step tests; clay autopilot fixes - #4
Merged
Merged
Conversation
Swing speed, friction, relief and house inertia, boom and arm flows per direction, and an optional command response (dead time and lag) were identified from PWRI's ZX200 measurements (Abe et al., ISARC 2025, doi:10.22260/ISARC2025/0054). Relief-based values are provisional. In clay the face planner also sets aside a strip whose pass carries less than half a bucket to the carrier, and set-aside strips are a fallback. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PWRI's constant-pilot bucket steps show how speed changes with angle, and the model's four-bar follows that curve when PWRI's bucket angle is read as the URDF angle + 0.622 rad (provisional, not confirmed by PWRI). With that reading the bucket takes 332 / 315 L/min to curl / dump, within 1 % of the measured speeds (Abe et al., ISARC 2025, doi:10.22260/ISARC2025/0054). Speed commands now ease onto the 5 cm moment-arm floor instead of jumping at it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Faster curl trims every cycle; sand and loam fill better per pass, clay worse. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…de at once A curl that loses more than a quarter of the load tore out a slab and is no longer learned from: one such curl in clay pulled the learned gain to 18 kg, and curling only an overfull bucket tore out more slabs. A poor strip is now set aside when the bucket reaches the carrier, before the next strip is planned, not one pass late. bench_haul.py logs every pass and can stop after PASSES=N passes. The 25-pass check against main doesn't show a gain yet: loam dumped 19 % less per pass, clay 5 % less (RESULTS.md). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
dtt-github
changed the title
Owner
Author
|
Updated: this PR now also has the bucket calibration (45f2f91, provisional angle offset of +0.622 rad, see RESULTS.md), the haul re-run with it (cce0787), and 3371b1c (curl-tear guard, earlier set-aside of poor strips, bench_haul.py per-pass log and PASSES=N). In the 25-pass check the two autopilot changes don't show a gain yet (loam −19 %, clay −5 % dumped per pass, single runs); they are being validated in a follow-up. |
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e and early set-aside reverted The pass count goes up only once the approach is reached, so 'swing back' and the approach were planned with the previous pass's strip: the house swung toward the new strip, back to the old one, then across again while penetrating, overshooting into the pit wall. That was the failing clay strip at +20 deg. _lane_pass() now aims both at the next pass's strip. Clay and loam dump about 11 % more per pass (25-pass runs, 3 each); clay over 1500 s loads 52-58 t instead of 43.5 t. The curl-tear rule lowered clay and loam per-pass mass, and setting poor strips aside early made no measurable difference, so both are reverted (A/B in RESULTS.md). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…d JSON The README still gave the spec swing (13.3 min^-1, 68.9 kN m) and the pre-calibration cycle (15-18 s, about 1 t per pass). It now gives the PWRI-calibrated swing, marks the swing torque and bucket offset as provisional, and the current pass times and payloads. haul_bench_lane_fix_long_* had Warp's startup output above the result line. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Autopilot: approach aims at the next pass's strip; tear rule and early set-aside reverted
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Calibrates the ZX200 model's swing, boom and arm hydraulics against PWRI's single-joint velocity step tests (1800 min⁻¹, eco mode, free air). The parameters were identified from PWRI's ZX200 measurements: Abe, Matsusaka, Endo, Yamauchi, Hashimoto, ISARC 2025, doi:10.22260/ISARC2025/0054. The recordings themselves are not in this repository.
Changes
command_response=Trueorros_server.py --command-response.RESULTS.md: new section "Calibration against PWRI's ZX200 step tests", with the measured fit and the benches.Fit with the command response on: swing, boom and arm speeds are within 4 % of measured, and delays within 0.05 s. The bucket is not matched yet; that is being fixed separately.
Benches (1500 s haul), before → now:
Passes take about 1 s longer. Clay's shortfall is still being investigated.
🤖 Generated with Claude Code