Repository navigation
The robot as it appears in the V2 videos: a 3D-printed head and pan/tilt neck with MG90S servos, the 2.42" screen, the Dayton CE32A-8 speaker in the head, custom head and base boards, and the Python brain with a dashboard, face packs and personalities.
Hardware:
- New 3D-printed neck and head with MG90S servos: pan geared up 38:30
(±95°), tilt ±85°, the 2.42" SSD1309 screen, the camera on its side in
the head, the Dayton CE32A-8 speaker and the amp in the head, power in at
the base. Replaces the off-the-shelf SG90 bracket. - Custom head and base boards in place of the perfboards (the board files
are in the hardware bundle; the connections are in the README).
Software:
- Clock and weather screen while Rocky is asleep, run by the robot itself.
- Dashboard (
server/launcher.py, http://localhost:8766/): live view,
settings saved inserver/settings.json, face packs, server log, and a
Start / Stop button for the brain. - Face packs: twelve emotions, four built-in looks, your own made in the
dashboard. - Personalities: make new characters in the dashboard from a short
description (name and wake word, Fish Audio voice, Fish's Drama 3 model
for sighs, laughs and whispers). Rocky stays the default. - Seeing: a picture with every question, sharp stills when he looks, and
a local model that finds faces so he can follow you. A wave wakes him. - Voice tuned for the CE32A-8 (bass cut at 300 Hz, peaks kept under the
amp's clean limit). - Language model: Claude Haiku 5.5 (was Haiku 4.5).
- Assembly mode (BOOT button) centres both servos for fitting the gears.
- Body language: small head moves with his emotion while he talks, and a
thinking glance; still while asleep, tracking, or taking a picture. - Computer control (off until you switch it on): he can search the web
and open sites on any computer, and on a Mac open apps, play music,
change the volume and run Shortcuts you allow.
docs/computer-control-windows.md
has ideas for the rest on Windows. - Robot link: a robot that restarts reconnects at once, the firmware joins
the strongest access point and adapts its speaker buffer, and the
connection token is checked more strictly.
The print files and board files are in the hardware bundle on my website (see the README). This release is a permanent snapshot: main continues as the latest version, and V1 stays available as v1.0. See CHANGELOG.md.