◆ PROJECT MILO — HUMANOID PLATFORM

Hello. I am being
assembled in public.

A bipedal, conversational robot built one subsystem at a time — Raspberry Pi 5 for the body, Jetson for the brain, RealSense for the eyes. Head is done. Hands are next. Everything I learn gets logged here, and every file ends up downloadable.

[ DROP HERO PHOTO OF MILO ]
head_v1 · 4:3 · 1600×1200

/ SUBSYSTEM STATUS

Head ACTIVE

2026-07-27 : Finally got shell had to get help from designer to create it for me, I added the camera bracket to hold RealSense D435if

50% COMPLETE
Hands ACTIVE

2026-07-20 -- Started Researching different hands for humanoids. 2026-07-25 -- Still In Design Phase, Setting up Fusion Project

10% COMPLETE
Vision SPEC'D

RealSense D435if streaming depth. No tracking loop yet.

5% COMPLETE
Compute SPEC'D

Pi 5 + Jetson split decided. Power budget open.

25% COMPLETE
Speech / LLM R&D

R&D

0% COMPLETE
Locomotion R&D

Bipedal walking. Hardest problem, deliberately last.

5% COMPLETE
Torso / Frame R&D

R&D

0% COMPLETE
Power R&D

R&D

0% COMPLETE

/ ANATOMY

▸ HEAD

Head + Sensor Pod

Printed shell, pan/tilt neck. Houses the Intel RealSense D435if for RGB + depth, mic array, and the speaker. First subsystem finished — everything else gets built to fit under it.

STATUS DONE

/ DEVLOG

Project start

I’ve decided to find out whether I can build my own humanoid robot. His name is Milo.

This is a passion project, and I’m documenting the whole thing as I go — the decisions, the dead ends, and everything I learn along the way. All of it gets shared.

The main constraint is cost. Almost every part gets manufactured at home: printed on a Bambu Lab H2C or milled on a Makera Air. Keeping it cheap isn’t a compromise, it’s the point — I want this to be something another maker can actually reproduce.

I’m not a designer or a mechanical engineer. Twenty-plus years of software engineering means I can make almost anything think. It does not mean I know what a moment arm is. Closing that gap is most of the project.

This is going to take a while. Some of the tools I need don’t exist yet, so part of the work is building the things that build Milo, before any of Milo exists. That’s fine.

Let’s get started.

/ ROADMAP

P0 · DONE

Head + sensing

Shell, neck gimbal, RealSense mounted and streaming.

P1 · IN PROGRESS

Arms + hands

Tendon-driven hand, forearm servo pack, first grasp.

P2 · IN PROGRESS

Perception loop

Track a person, follow with the neck, no jitter.

P3 · QUEUED

Voice + conversation

STT, LLM, TTS on the Jetson. Latency under a second.

P4 · QUEUED

Torso + power

Frame, compute bay, battery and rail isolation.

P5 · QUEUED

Legs + walking

Bipedal balance, then flat-floor walking. The big one.

/ BILL OF MATERIALS

Raspberry Pi 5 (8GB)
body controller
IN HAND
$80
NVIDIA Jetson
perception + LLM
IN HAND
$249
Intel RealSense D435if
RGB + depth vision
IN HAND
$334
Neck pan/tilt servos
head articulation
IN HAND
$45
Hand servo pack
finger actuation
TESTING
$120
Mic array + speaker
voice I/O
TESTING
$60
Leg actuators
locomotion
TBD
Battery + power board
energy
TBD
Filament + hardware
everything else
ONGOING
$180

/ FILES & DOCS

STL / STEP
Head shell v1

Faceplate, rear shell, sensor pod bracket.

◦ PUBLISHING SOON
STL / STEP
Hand linkage v2

Fingers, palm, forearm servo mount.

◦ IN TESTING
SOURCE
milo-firmware

Pi 5 servo + sensor loop, serial protocol.

◦ NOT YET PUSHED
SOURCE
milo-perception

Jetson vision pipeline and tracking loop.

◦ NOT YET PUSHED
PDF
Wiring diagram

Power rails, servo bus, sensor loom.

◦ DRAFT
CSV
Bill of materials

Every part, supplier link, price paid.

◦ DRAFT

/ CHANGELOG

  • v0.4 Depth stream running on Jetson
  • v0.3 Compute architecture locked (Pi 5 + Jetson)
  • v0.2 Head assembly complete
  • v0.1 Scope defined, parts ordered

/ ABOUT

Milo is a humanoid robot I'm building from scratch, and this is the lab notebook. Twenty-plus years as a software engineer means I can make almost anything think. It does not mean I know what a moment arm is. Closing that gap is most of the project.

Almost every part gets manufactured at home — printed on a Bambu Lab H2C, milled on a Makera Air. The constraint is deliberate: keep this cheap enough that anyone with the same two machines can follow along and build their own. Some of what I need doesn't exist yet, so part of the work is building the tools that build Milo.

Everything gets documented — the decisions, the part numbers, the print that warped at hour nine. Every file ends up downloadable. This is going to take a while, and that's fine. Let's get started.