Mechanical

Material and mechanism selection, sourced. CAD still to come

Open on GitHub Mở trên GitHub

Snapshot of this document in the GEMs repository. The repository itself is the current version. Bản chụp tài liệu này trong repo GEMs. Bản đang chạy nằm ở chính repo.

Material and mechanism selection against the specification. Every figure is sourced; §6 lists the references.

No CAD yet. What is fixed here is what the CAD will be made of and what the joints will be made from — the decisions that have to precede geometry.

1. Structure

Selection: carbon fibre composite for long members, 7075-T6 at joints and load-introduction points.

This is not a preference, it is what the mass loop leaves room for. The actuator finding in ../electrical/ shows f_act is under severe pressure — nothing below 80 Nm/kg keeps it inside the assumed range. If actuators are going to sit at the top of their band, structure has to sit at the top of its own, and a 40% saving against 7075 is the difference between a loop that closes and one that does not.

Aluminium stays where composite is poor: bolted joints, bearing seats, anywhere load enters a member through a small area. Composite carries length; metal carries concentration.

f_str is not closed by this. Material fixes the density and the modulus; the mass fraction depends on section sizing, safety factor and joint design, none of which exist yet. Spec 02.2 leaves it at 0.25–0.35 and it stays there — but the choice above is what makes the lower end reachable rather than aspirational.

2. Reducers, by joint class

The arms are the interesting case, and the specification decides it. Harmonic drives are the industry standard for robot arms [6] and are lighter than cycloidal at the same ratio — but at high ratio they are effectively non-backdrivable. A stalled arm cannot yield: contact force becomes whatever the controller commands, with nothing mechanical setting a ceiling.

Spec 01 puts safe human contact in the core group. QDD at 6:1–15:1 keeps the arm backdrivable, which makes compliance a property of the mechanism rather than a promise made by software that can fail. The cost is torque density, and §1 of the electrical selection shows there is very little of that to spare — so this is a real trade made deliberately, not a free choice.

Cycloidal reducers are heavier and bulkier with more parts [5], which is acceptable in the legs and not in the wrists.

3. Protection

The group 5 standard requires handgun rounds, blades and everyday impact together, and bullets and blades defeat armour by different mechanisms — a blade tip passes between soft fibres where a bullet does not. A single body facing both needs a layered package.

Against 1.6 m² of skin area at 1.65 m, at the coverage bands of spec 02.4: 4.8 kg at 50% coverage and 6 kg/m², up to 11.5 kg at 80% and 9 kg/m².

Armour is non-scaling mass. It enters m_ext and is multiplied by γ, so at γ = 2.6 the difference between 50% and 80% coverage is ~3 kg of armour and ~7 kg of body. With f_act already under pressure, coverage is the cheapest place to buy margin back — and the most visible thing to lose.

4. Shell layers

Spec 04.5 divides the shell into three layers that cannot substitute for one another: sense-and-heal, variable stiffness, and load-and-fire. All three are non-scaling mass competing with armour inside the same 15–25 kg budget.

Areal densities for the first two remain ⟦IMPL⟧ — no reliable published figure was found, and inventing one would put a number into the mass budget that nothing supports.

5. Open

6. References