Skip to content

armgit / moss

Public Imported from GitHub
0

MOSS: a small, mostly 3D-printed litter-picking rover. Cleaner streets, one maker at a time.

Python
4.5 KB
Raw Blame

Changelog

Hardware revisions. Dates are when it was real, not when it was planned.

  • 24 October 2026, the pad drives from the battery switch. Power on and the rover is drivable once its stack is up: tools/moss-teleop.service runs the launcher at boot and retries until the pad and the board answer. No deadman: a small tracked rover at walking pace does not need a held button, a stick out of its deadzone drives and centred sticks stop (0.5 s of zeros, then silence); the neutral-first trust gate and the firmware's 300 ms stop stay, MOSS_PAD_DEADMAN=r1 in robot.env brings the held button back. The pad layout is picked from the device's kernel name (MOSS_PAD_LAYOUT=auto): ds4 joins android and xinput, for a 2.4 GHz receiver that enumerates as a DualShock 4 (a GameSir Nova 2 Lite on its receiver, DInput mode); MOSS_PAD_INVERT_Y and MOSS_PAD_INVERT_TURN flip a stick whose sign is the other way round. Sound from the rover itself: a MAX98357A amplifier on the Jetson's I2S2 pins and a wired speaker (Software.md), the sound pack normalised to −1 dBFS peak (it sat at −12 dBFS). USB: a hub carries the pad receiver, the wrist camera and the lidar; the RealSense, the motor board and the servo bus go straight to the Jetson. Six minutes of driving indoors on one charge: driver and motors cool to the touch after it.
  • 11 October 2026, dimOS software moves into this repository (moss_dimos): base, lidar odometry, map, planner, explorer, gamepad, recorder, flight check, launcher, cockpit. dimOS itself stays an outside dependency, installed from git (Software.md).
  • 10 October 2026, first drive at the gamepad. The V0.5 rover, arm mounted, driven on the floor from a gamepad through dimOS: forward, reverse, turns, pivots. What it took and what it showed:
    • Firmware qtpy-mdd3a-v3 on the QT Py ESP32-S3 (firmware/MOSS_QTPY_MDD3A/): the 25 % commissioning limit is gone, full stick is 100 % PWM (at 25 % the loaded tracks did not move at all); left track direction fixed in the firmware, encoder counts follow; ramp UP DOWN sets the rise and fall slopes live; a build switch for the Cytron MDD10A (PWM + DIR) next to the MDD3A build; the board names its driver in the banner and the telemetry. Host test updated.
    • Teleop: the base runs as a dimOS blueprint (moss-gamepad, now in this repository's moss_dimos): deadman on the pad, twist to track percent, keepalive inside the firmware's 300 ms deadman, tank steering (past half turn stick the inner track reverses), one launcher with gates (pad, board, firmware, battery) and a telemetry line every two seconds while driving.
    • Seen under load: the board applies 100 %, the 3S battery drops from 11.6 V at rest to 10.8 V in a pivot. Verdict on the drive line: the MDD3A (3 A per channel) and the 56:1 motors leave no headroom for ramps. Next on the drive line: Cytron MDD10A (10 A per channel) and JGB37-520 motors at 110 rpm (higher ratio), both not installed yet; the firmware build for the MDD10A already exists.
    • Tracks: the double tensioner holds, the bearings keep mechanical friction low; the TPU tracks may need running in, or print settings that make them more supple.
    • In CAD, no part printed or fitted: hull R19 with an integrated mount for the MDD10A (rotated hole pattern 55.88 x 78.74 mm, four M3 x 6 into 2.7 mm pilot holes, no inserts), charging jack moved to clear the larger board, the motor-side drive-spacer passages opened to 9.5 mm; transmission R12-C1 (two drive-shaft spacers that were missing, 3.5 mm, PETG) and R12-C2 (matching screw-head recesses on the inner frames and inner tensioner plates, so all four tension screws become M4 x 35). CAD checks only; the production plates and the BOM are not reconciled to it, and nothing is frozen.
    • Still open on V0.6: fan spacing and clip retention next to the motor brackets, inserts for the LiDAR mount, upper cover and arm platform, the LiDAR collision, one consistent hole pattern per wheel type.
  • V0.6 (October 2026): modular compute and perception plates (Jetson or Raspberry Pi; RealSense or Orbbec Gemini), XT30 power block, 12 V to 5 V converter for the Pi, hull R7, tracks 70 teeth at 10 mm pitch.
  • V0.5 (1 October 2026): redrawn on the real robot; one-piece hull, two tensioners per track, mounts for every board, magnets for the bin. 107 printed parts.
  • V0.4 (24 September 2026): revised hull, screw-terminal controller, one dual motor driver.
  • V0.3 (22 September 2026): first prototype. First drive on 22 September, arm and gripper on 26 September.