Conversation
The hand-coded translation/rotation formulas caused large spurious outputs on Z, Rx, Ry during pure XY translation (confirmed via log analysis). Replaces the formulas with a 6x9 calibration matrix (Config::DECOUPLING_M) fitted from per-DOF calibration movements using least squares, so each axis responds only to its intended physical input. Changes: - MotionController: matrix multiply replaces hand-coded formulas - Calibration.h: generated file holding the matrix (included by Config.h) - TelemetryController: now emits 9 baseline-subtracted sensor deltas (s0x..s2z) alongside motion values, for use by calibration tooling - monitor.py: logs sensor deltas + motion in tab-separated format with header - calibrate.py: fits the 6x9 matrix from 6 labelled log files, writes Calibration.h - run_calibration.sh: walks builder through each DOF movement end-to-end Other builders run ./run_calibration.sh then reflash to generate their own Calibration.h matched to their specific hardware assembly. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Documents the purpose of the sensor decoupling matrix, step-by-step calibration instructions, and how the least-squares fit works in README.md. Includes the updated run_calibration.sh script and regenerated Calibration.h. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…el bugs - Remove double dead-zone: hardZero() on output was zeroing motion while the lowpass filter ramped up from 0, causing a sticky delay at the start of every movement. Output is now just clampf(filt_[i], ±AXIS_LIMIT). - Delete unused GAIN_T, GAIN_R, SIGN_AXIS from Config.h (pre-decoupling-matrix leftovers with no callers). - Default ENABLE_TELEMETRY to false so production builds don't ship with serial output and a 200ms startup delay. - Make readRaw() return bool and check TLx493D read success; IdleState and updateCalibration now skip the frame/sample on I2C failure. - Replace millis()==0 sentinels with explicit bool flags (calibrationSampleStarted_, spinnerStarted_, bothHeldActive_) in SensorController, LEDController, and InputController. - Move baseline subtraction for telemetry into TelemetryController::publish() to eliminate the redundant delta loop in IdleState. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds spacemouse-buttons.py, a hidraw-based listener that fires xdotool key sequences when a button is pressed, matched against the active window title. Reads from /dev/hidrawN directly (non-exclusive) so it coexists with spacenavd which holds an exclusive grab on the evdev node. Key design decisions: - buffering=0 on hidraw open — preserves one-report-per-read kernel guarantee - SIGHUP reloads config without restarting the service - Restart=always in the service so it recovers automatically after device replug - button-map.conf ships with common CAD app mappings (FreeCAD, SolidWorks, Inventor, Rhino, Fusion, Blender) with a * fallback; user config at ~/.config/spacemouse/button-map.conf is never overwritten on reinstall linux-setup.sh updated to install xdotool, write udev rules for both input and hidraw subsystems, and deploy the button mapper as a systemd user service alongside the existing spacenavd setup. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
… needed - HID descriptor gains report ID 4 (feature: 9 sensor-delta floats) and report ID 5 (output: 6×9 calibration matrix). TinyUSB callbacks route GET/SET report to HIDController. - MotionController.setDecouplingMatrix() lets the matrix be updated at runtime without reflashing. Falls back to compiled-in Calibration.h. - CalibrationStore reads/writes the matrix to LittleFS flash so it survives power cycles. Loaded on boot in main.cpp setup(). - IdleState exposes baseline-subtracted sensor deltas each loop cycle (for monitor.py to poll) and applies any incoming matrix from the host. - monitor.py rewritten: polls HID feature report via HIDIOCGFEATURE ioctl instead of reading serial. Reads motion axes from HID input report ID 1. Same log format, so calibrate.py is unchanged. - calibrate.py sends the fitted matrix to the device via output report ID 5 (write on hidraw) after fitting, giving immediate effect without reflash. - run_calibration.sh no longer mentions serial ports or firmware builds. - platformio.ini collapsed back to one build environment. - mouse_monitor.log untracked (was accidentally committed earlier). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…kflow - Add Linux Setup section covering linux-setup.sh, spacenavd, and the button mapper (config, reload, debug commands). - Update Calibration section: remove pyserial requirement and reflash step; explain that the matrix is sent to the device over USB and persisted to LittleFS flash. - Expand "How it works" to describe HID feature report polling, matrix fitting, and the output report + LittleFS persistence path. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When hardZero() was removed in bb5b9c8 to eliminate the start-of-motion lag, it inadvertently broke standby. The old code only set motionActive_ when filt_[i] had ramped above the dead zone; sensor noise spikes could briefly push y[i] above dead, but filt_[i] would snap back to 0 on the next frame and never accumulate, so motionActive_ stayed false at rest. With out[i] != 0 as the check, any noise spike that briefly moved filt_[i] off zero would set motionActive_ = true, keeping lastActivityMs_ fresh and preventing the 2-minute sleep timeout from ever firing. Fix: check fabs(filt_[i]) >= dead instead. Since filt_[i] snaps to 0 whenever y[i] drops below dead, noise cannot accumulate in the filter — only sustained motion builds filt_[i] above the threshold. The HID output (out[i] = clampf(filt_[i], ...)) is unchanged, so there is no regression to the start-of-motion responsiveness fix. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Every caller (real-time motion, baseline calibration, and the HID feature report calibrate.py reads) was seeing a single unsmoothed I2C sample, which fed noise straight into the dead-zone comparison and into calibration data collection. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…itives Hold both buttons 10s to reboot into the UF2 bootloader for drag-and-drop reflashing without opening the case. Hold the right button alone for 3s to cycle the idle LED color, persisted in flash across power cycles. Also adds setPixel/setPixelGroup/setTwoGroups/setPixelGroupOnBackground to LEDController, used by the live direction indicator and diagnostic HID LED commands. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
MotionController now computes a 3D point and orientation from the 3 sensors' real PCB mounting positions instead of mapping raw sensor deltas straight through a single fitted linear matrix. A second, ridge-regularized 6x6 matrix (GEO_M, GeoCalibration.h) then decouples the crosstalk this introduces (the short sensor baseline amplifies any Z-asymmetry into a large apparent tilt angle), and a per-axis gain lets translation axes compete fairly against the rotation crosstalk they induce. Ported from interactive_calibrate.py, where this was developed and verified live. Also: - Adds diagnostic HID output reports (6/7) for host-driven LED color/ pixel control, and wakes SleepState to apply them or a pushed calibration matrix instead of silently dropping them. - Rewrites the idle-state LED direction indicator to match lightpattern.md exactly: idle cyan, push/pull full-ring red/green, tilt shows 2 green LEDs on a cyan ring, slide shows a green/red half-split, twist shows a blue dot circling a cyan ring. All active indicators blink/animate at a rate proportional to magnitude instead of a fixed interval. - Fixes board_build.filesystem_size (was unset, silently giving LittleFS a 0-byte partition so calibration/LED-color persistence never actually worked). - Swaps which physical gesture drives HID Y vs Z (slide fwd/back vs push/pull) — spacenavd/FreeCAD treats HID Y as dolly/zoom, which made slide fwd/back zoom instead of pan. - GEO_AXIS_GAIN[TX] and GEO_AXIS_GAIN... TZ (existing AXIS_GAIN) are negative to fix inverted CAD view motion found in live testing; the LED direction logic compensates so indicators still match the physical motion. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Self-paced host tool (run directly in a terminal, not through an AI assistant) for tuning and verifying the geometric motion model live: menu 1-6 captures 5s of motion per axis, 'g' fits the GEO_M decoupling matrix from those captures, 'c' runs the existing linear-matrix calibrate.py, +/- tune the 3D-point scale factor live. Shows raw per-sensor deltas, the fitted 3D point/orientation, and a plain-English "THE MOUSE THINKS: ..." line reflecting the same logic as the firmware's LED indicator. geo_matrix.json is fit from one physical unit but included as a starting point for similarly-built units, since the sensor mounting geometry is fixed by the PCB design. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Geometric motion model, LED direction indicator, and misc firmware fixes
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
2 participants
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.
Summary
This started as a standby-timeout fix and grew into a larger rework of how motion is computed and displayed. Opening it here in case any of it is useful upstream — happy to split into smaller PRs if that's easier to review, or if parts don't fit your direction just take what's useful.
motionActive_now checksfabs(filt_[i]) >= deadinstead ofout[i] != 0, restoring the LED-off timeout without reverting the start-of-motion responsiveness fix from an earlier commit.interactive_calibrate.py). The old linear matrix is still fit/pushed/persisted as a compiled-in fallback but no longer drives live output.lightpattern.mdfor the exact spec): idle cyan, push/pull full-ring red/green, tilt shows 2 green LEDs on a cyan ring, slide shows a green/red half-split, twist shows a blue dot circling a cyan ring — all active indicators blink/animate at a rate proportional to motion magnitude.board_build.filesystem_size(was unset, silently giving LittleFS a 0-byte partition — calibration and LED-color persistence never actually worked).interactive_calibrate.py, a standalone host tool for live-tuning/verifying the geometric model, plus a fittedgeo_matrix.jsonas a starting point for similarly-built units (sensor mounting geometry is fixed by the PCB design, so it should be a reasonable starting fit for other units of this board).Caveats
GEO_OUTPUT_SCALEin particular) were tuned live against one physical unit and will likely need retuning on others.spacemouse-emulationwork I saw open here — different philosophy (geometric model vs. per-unit linear fit), not a drop-in replacement for that.Test plan
pio run -e seeed_xiao_rp2350)