Every driver must normalize its device's native scale into ROS REP-103
right-handed CCW (0 = front, + = left). Spells out the three stackable
device quirks to absorb — shifted 0° reference, clockwise scale,
upside-down mounting — and warns that a CW scale decoded as CCW mirrors
the world invisibly on single-lidar devices: verify handedness against
the real room, not just against motion.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Slamtec devices count their angle clockwise viewed from the top
(Interface Protocol spec; the official rplidar_ros node mirrors the
angles for the same reason), but the decode used the raw ascending
angles as CCW for a right-side-up unit and mirrored them for an
upside-down one — exactly backwards. Field-verified 2026-07-23: a
right-side-up C1 produced a left-right mirrored sweep against a
verified right-handed reference lidar.
Now: right-side-up mirrors (angle' = 2π − raw, nodes walked backwards);
upside-down uses the raw ascending angles, because the physical flip
already reverses the apparent rotation. The mountedUpsideDown flag
finally carries its true physical meaning.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The LR-1BS5's angle scale runs clockwise viewed from the top, so
decoding it as CCW mirrored the whole world left/right — undetectable
with a single lidar (the SLAM map is self-consistently mirrored and
heading-vs-motion checks pass, since a mirror about x preserves
'ahead'). It surfaced when a second, right-handed lidar disagreed
(doubled walls), and the finished map came out mirrored versus the
actual room.
device_deg() now negates the raw angle BEFORE the model's 0°-reference
offset (LR-1BS5: out = 180 − raw), driven by model_angles_clockwise().
Evidence: field-verified on OLELR-1BS5 (2026-07-23), and DF Automation's
production ROS driver (github.com/dfautomation/ole2d, decoder.cpp)
walks the device array backwards with the comment 'reverse, laserscan
is anticlockwise' — its packet layout is exactly our Family B, so
LR-1FMI (same protocol) is listed too. Family A/C stay CCW until a
unit is verified against a real room.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Field-verified on an OLELR-1BS5 unit: with offset 0 the SLAM pose moved
correctly but the estimated heading came out 180° from the direction of
travel (the classic rotated-sensor-frame signature). The 1BS5 numbers
its angles from the rear like its LR-1F/LR-1FMI siblings, so it gets the
same angle_offset_deg = 180 so that driver output 0° = ahead.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The repo is now self-contained: driver_rplidar compiles the SDK from
third_party/rplidar_sdk (Linux sources only, teardown delete[] bug fixed
in place), so a plain cmake configure needs no SDK path and no probing —
the XLIDAR_RPLIDAR_SDK_DIR option and external references are gone.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
DeviceConfig now carries an optional transport (serial/udp/tcp) instead of
the ESPE-only use_udp bool. Plugins validate it in create_driver_instance:
a fixed-transport driver configured with the wrong transport fails open()
with InvalidConfig (via InvalidConfigDriver — the plugin ABI forbids
returning nullptr) rather than silently ignoring the setting. Selectable
drivers (ESPE) switch TCP/UDP through the same field. config.json
load/save round-trips "transport" for every transport, including serial,
and migrates legacy use_udp:true entries.
Diagnostics drops the per-vendor accessors (espe_fault, rplidar_fault,
monitor_fault, sick_error, pollution_*, contamination_*, manipulation) for
one common shape: a list of DiagnosticIssue{severity, code, detail} with
cross-vendor codes, plus a raw map of vendor passthrough values and
to_json() for hosts that prefer a string. Vendor bit decoding now lives in
one place (decode_diagnostics); has_fault/has_warning/healthy keep their
meaning, so is_ready()/wait_ready() are unchanged.
Also: README regains the model/protocol and ExtraInfo tables lost in the
lidarlib->xlidar refactor (verified against current code), and the empty
xlocd/ tree left by a stray sync run is gone.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The driver targets the SICK TiM 5xx/7xx family over CoLa-A; "code" described
neither the devices nor the protocol. driver_id stays sick_tim_driver.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
parse_family_b() dùng sai hệ số góc 0.25°/LSB; theo spec Olei chính hãng
(Olei.LidarSensor/LidarDataBlock.GetAngleDegrees) AngleRaw là 0.01°/LSB.
Sai 25× khiến điểm bị gán nhầm góc → một phòng bị bôi thành vòng tròn trên
RViz. Đã verify với thiết bị thật OLELR-1FMI: sau khi sửa ra 2400 điểm/vòng,
0–359.9°, đúng hình học môi trường.
- Đổi hệ số góc 0.25° → 0.01° trong parse_family_b().
- Bỏ qua block invalid (AngleRaw >= 0xFF00) theo spec.
- Dò ranh giới vòng quay PER-POINT thay vì per-packet (một gói có thể chứa
>1 vòng), tránh gộp nhiều vòng vào một scan.
- Thêm model LR-1FMI (360°, 0.01°/LSB, ~2400 pts/rev) vào bảng model +
kModelTable, đặt "1FMI" trước "1F" để khớp đúng chuỗi tên.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
