Skip to content
Menu

Roadmap

Broad roadmap, narrow active scope.

A wide map costs almost nothing to keep. Doing many things at once costs everything. ME holds the first and refuses the second.

How execution works

01

Broad roadmap

The map covers computing, robotics, energy, space, manufacturing, materials, medicine, transportation and frontier physics. Keeping the map wide is cheap. It is what lets one branch borrow from another.

02

Narrow active scope

Only a few projects are active at once. Planning stays shallow until a branch genuinely needs deeper design, rather than pretending every subsystem needs a five year plan today.

03

Milestone gates

Each project advances through small milestones that can be individually demonstrated. A milestone is done when it runs, not when it compiles.

04

Projects earn expansion

Scope is granted, not assumed. A project widens after it has shown a working milestone, which is why the active list is short and the map is long.

05

Shared infrastructure

ME OS, Carl, the research tools platform and manufacturing are shared across products, so work done once is reused instead of rebuilt per project.

Selected current milestones

A milestone is done when it runs and can be demonstrated, not when it compiles.

  • ME OSComplete

    M1: boot proof

    Boot in a virtual machine to a black screen reading IF YOU SEE THIS IT WORKED. Nothing else. A freestanding x86-64 kernel boots over UEFI, draws the line, and does not crash. Checked automatically by inspecting the framebuffer.

  • ME OSComplete

    M2: keyboard input

    The boot message stays where it was and a line below it reports the last key pressed. Verified in QEMU by injecting a key press and inspecting what was drawn. Not yet run on a physical machine.

  • ME OSComplete

    M3: draw a rectangle

    One static filled rectangle, with the boot message and the key line untouched. Checked automatically: the captured framebuffer must contain a solid rectangle of the expected size, centred and clear of the text, alongside both lines of text. Framebuffer clipping is also checked on the development machine with guard regions around a fake framebuffer.

  • ME OSComplete

    M4: mouse cursor

    A cursor drawn on the framebuffer that follows the mouse, keeps its shape, and stays on screen. Checked automatically: the emulator moves the mouse and the captured framebuffer must show the cursor moved by exactly that much, with the message, key line and rectangle all untouched.

  • ME OSComplete

    M5: move the rectangle

    The rectangle crosses the screen at sixty pixels a second and turns around at each edge. It moves at a rate rather than at whatever speed the machine runs its loop, because the kernel now reads a hardware timer. Checked automatically across four captures: the rectangle is whole, on screen, and somewhere different each time.

  • ME OSComplete

    M6: basic arithmetic

    Type a sum and press enter, and a line above the boot message shows the answer. Addition, subtraction, multiplication, whole number division and powers, with the precedence they have on paper. Overflow, division by zero and fractional powers are refused and shown as an error rather than producing a wrong answer, which matters in a kernel with no interrupt table to catch a fault.

  • ME OSComplete

    M7: conditionals

    One conditional expression on the same line: IF one value compared to another THEN this ELSE that, with =, < or > as the comparison, and a sum allowed in all three places. Checked by typing both a condition that holds and one that does not, and reading the two different answers off the framebuffer.

  • ME OSComplete

    M8: variables

    A value can be given a name and used on any later line. Eight names fit, each an uppercase letter and up to three more characters. Reading a name that was never set is an error rather than zero, because a typo that quietly reads as zero gives a wrong answer and says nothing about it. One table, no scope, and nothing survives a reboot.

  • ME OSNext

    M9: keyboard controlled rectangle

    Keys move the rectangle, which until now has only moved on its own. After it: wrapping at the screen edges, then picking the rectangle up with the pointer.

  • HoloprojectorComplete

    M1: rotating pyramid

    A pyramid in a 3D simulator, rotating on elapsed time, with pause, resume, reverse, axis, speed and reset, all driven through the command layer rather than the renderer.

  • HoloprojectorComplete

    M2: multiple scene objects

    Several independent objects in one scene, each with its own transform, visibility and rotation, plus selection and per object commands. Still a simulator. No display hardware exists.

  • HoloprojectorComplete

    M3: pointer interaction

    A device neutral pointer: hover, select, grab, drag and release. A pointer sample is a ray plus a pressed flag, so a mouse, a holo pencil, a bracer or a hand tracker could each produce one. Only the mouse works; the others are named stubs that refuse to poll. Dragging goes through a command that safety checks, so a drag out of the display volume is refused.

  • HoloprojectorComplete

    M4: simulated Holo Pencil

    A second pointer source, simulated in software: a tip in the display volume that the keyboard walks around. It selects, grabs, drags and releases through the same controller, commands and safety checks as the mouse, and a drag begun with one source can be finished with the other. No physical pencil exists.

  • HoloprojectorComplete

    M5: bracer input adapter

    An adapter shaped like a wearable: a hand pose in its own tracking space, an alignment that maps it into the scene, and a grip. It can also lose tracking, which lets go of whatever was being held instead of leaving it attached to a hand nobody can see. None of that needed a change in the interaction layer, the commands or safety. No bracer exists, and nothing reads or models a sensor.

  • HoloprojectorNext

    M6: richer Carl adapter

    Broader phrasing and multi step instructions through the same command path, still model agnostic and still offline.

  • Employee BracersPlanned

    Hand tracking feasibility

    Establish whether inertial and optical sensing, and later surface electromyography, can resolve hand and finger intent accurately enough for holographic interaction, before any wearable is designed.

  • Research ToolsPlanned

    Platform design

    Define the shared simulation, measurement and experiment tracking layer other branches would build against.

  • EnergyPlanned

    Smart DC bus study

    Study a shared DC distribution bus for a lab bench or factory cell, with monitoring and protection built in. Paper study only. Nothing has been built or energised.

  • ManufacturingPlanned

    First production proof of concept

    End to end automated production of one simple non safety critical automotive clip, using additive manufacturing, with camera and weight quality control and full traceability for rejected parts. Nothing has been built.