PHASE 1

Software stack

Local, low-latency, killable. High-level conversation may hybridize. Motor authority stays local under a safety supervisor. The Sense Head nodes are MIT and public.

MIT ROS 2
Source: github.com/Pitchfork-and-Torch/HumanBotty ยท package software/
Open the Sense Head code Repo

Clone

git clone https://github.com/Pitchfork-and-Torch/HumanBotty.git
cd HumanBotty/software
# tests (no ROS required):
python -m unittest discover -s test -v
# on a robot with ROS 2 Humble or Jazzy:
# colcon build --packages-select humanbotty_sense_head
# ros2 launch humanbotty_sense_head sense_head.launch.py

On-robot

  • Ubuntu (Jetson BSP) or Ubuntu / Pi OS
  • ROS 2 Humble or Jazzy (pick one, stick)
  • Python 3 nodes; brand SDKs for camera and servos
  • safety_supervisor always on the motor path

Topic contract v0

/humanbotty/camera/color/image_raw
/humanbotty/camera/depth/image_raw
/humanbotty/imu/data
/humanbotty/audio/vad
/humanbotty/head/joint_states
/humanbotty/head/joint_command
/humanbotty/safety/estop
/humanbotty/status

Intelligence bridge

ROS 2 topics
  -> humanbotty_bridge
      -> local policies (look-at-face, look-at-sound)
      -> optional LAN link to a desktop agent session
      -> NEVER bypass safety_supervisor

Phase 1 local policies (proudly simple)

  1. Face detect, pan-tilt toward centroid
  2. Loud sound, orient
  3. Idle micro-saccades (limited)
  4. Voice I/O via your agent stack when ready