Plone Mraz

Energy

Sources, the six levers, state levels, docking, quiescent power and sleep ceiling

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.

Every trade in this specification is ultimately a division of a kilowatt-hour budget. This chapter states the budget, the measures that stretch it, and the ceilings that cannot be stretched.

3.1 Source

Two configurations, declared by what the cell must deliver rather than by its chemistry:

Configuration Density Trade
Durable ~400–500 Wh/kg Good cycle life; thermal safety from the cell where it is solid-state, from the pack where it is not. Recommended as the design target
Ceiling ~700–1100 Wh/kg (solid-state Li-S) Maximum endurance. Cycle-life debt: presently ~100 cycles — the pack becomes a consumable

What the durable tier requires. Density of 400–500 Wh/kg at cell level, which is the e of the mass loop (02.2); a cycle life that keeps the pack from becoming a consumable; the discharge rate that 02.7 prices; and an abuse test the pack must pass. The figures for the last three are ⟦IMPL⟧: they are set when the cell is chosen (plan D-5), not invented here.

Solid-state is preferred, not required (Plone, 2026-11-05). It is preferred for a body in human contact: a solid electrolyte has no liquid to leak or burn. But this is a reading of present technology, not a lock. As of 2026-11, the durable density can be bought only as silicon-anode lithium-ion, with a liquid electrolyte; solid-state cells at that density stand at stack level, not on sale (hardware/electrical §3). Either fills the tier. Where the cell is not solid-state, the pack carries the safety the cell does not: a crush zone between shell and cells, and a venting path for thermal runaway away from the head (plan, pack location). When solid-state cells at this density can be bought, the preference decides.

Sourced, for positioning the three maturity tiers:

The pack is fixed in the unit: a part replaced in service, not during operation (Plone, 2026-11-05). Endurance beyond one pack comes from the dock (3.5); presence beyond one unit's endurance comes from other units, and which unit runs, and when another takes over, is the Main Controller's decision (01). As the bill of materials stands, the always-on rail and the supercapacitor sit in the power assembly with the cells, so removing the pack in service cuts the always-on rail (3.6); what must survive a service removal is ⟦IMPL⟧.

Peak-burst supercapacitor: optional. Fitted when the mission needs sudden high-power motion (catching, jumping, fast reflex); omitted when lightness matters more. Its role is millisecond-scale discharge without sagging the pack. See 02.7 for why this option decides more than its name suggests.

3.2 Six energy measures

# Measure Mechanism Effect
1 State-dependent gating Actuators tense only when needed; full-rate sensing only on attended channels; edge compute clocks down when still Strongest. Draw varies ~1:8 between idle and peak
2 Regeneration on deceleration Recover on limb lowering and braking up to ~30% (sourced)
3 Source tiering Main pack for motion, supercapacitor for bursts, trickle charge underneath Each tier does what it is good at
4 Mechanical joint locks Latch joints when holding a pose; actuators release Removes the cost of standing still
5 Actuator operating point Series-elastic elements store and return gait energy Cuts the static-hold penalty
6 Tiered sleep with delegated watch See 3.3 The only measure that attacks total scheduled energy

3.3 Tiered sleep

The body drops to quiescent power on a cycle: actuators off with the body resting on a mechanical support, full-rate sensing off, heavy compute off. A very low-power always-on wake-up circuit and wake-on-event remain.

Delegated watch. While the body sleeps, the off-board compute of the controller stays awake, observing through other sources — environmental devices, other units, signal networks. The body sleeps without the system going blind. This is an architectural advantage biology does not have, and it is available only because of the on-board / off-board split of 01.

# Constraint Content
1 Depth against wake latency Deeper sleep, slower waking: tens of ms (light) to seconds (deep). Multiple levels are offered; which one is ⟦CTRL⟧
2 Sleep posture is a physical constraint Switching actuators fully off requires a passive structure to carry gravity. Deep sleep needs support — it is not available while standing unaided
3 Quiescent power is above zero The always-on wake-up circuit still draws. Sleep is very long but not indefinite. See 3.6 for what actually bounds it
4 It cuts the integral, not the peak This measure reduces total energy over a scheduled cycle, not instantaneous power

3.4 Four state levels

Level Source Actuators Sensing Character
1. Active Battery Running Full, under gating All six measures apply
2. Free sleep Battery (floor) Off, body resting Always-on wake-up circuit only Maximum saving; the off-board compute watches
3. Rest-and-charge External Light support Spatial sensing retained Resting without going blind, charging without going dead
4. Deep sleep on dock External Off Minimal Fast full charge plus maximum rest; suits standby units

Level 3 is the state characteristic of this architecture. On external power, the constraint "switch things off to save the battery" disappears — nothing is drawing from the pack. The body recovers energy while maintaining observational presence. Organisms cannot do this.

Even where power is free, the dock should still carry most of the body's weight — not for the electricity, but for heat (actuators holding a pose get hot) and mechanical wear.

3.5 Dock

Form Role Gravity path Trade
Seat or couch (default) Active unit resting and charging Supported from below, in compression Low centre of mass, natural posture, good sightlines, fast to rise
Load-bearing hanger Standby unit in storage Suspended from above, in tension Less floor area, more units stored; smaller charging contact, poorer sightlines

Both charging methods are provided:

Which method is used when: ⟦CTRL⟧.

3.6 Quiescent power and the sleep ceiling

Components of quiescent power:

Component Power Note
Sub-GHz wake-up receiver 3–30 µW Sourced — published designs from 305 nW to ~6 µW; Wi-Fi-based variants ~30 mW
RAM retention and the root-of-trust monotonic clock ~10–50 µW ⟦IMPL⟧
Low-power beacon (06.5) < 1 mW
Spatial sensing, passive mode ~18 µW Sourced
Spatial sensing, full CSI, duty-cycled ~50–200 mW ⟦IMPL⟧ — the expensive term

But the always-on wake-up circuit is not the dominant term. Cell self-discharge runs 1–3% per month (sourced; automotive standards require under 2%/month, and solid-state is expected to be lower but has yet to demonstrate it). On the 3.9 kWh pack of the 2-hour reference body (02.7):

Self-discharge Equivalent power
1%/month 54 mW
2%/month 108 mW
3%/month 162 mW

A deep-sleeping always-on wake-up circuit costs tens of µW — about four orders of magnitude below self-discharge.

Sleep duration at 2%/month, linear approximation:

Mode Total floor Duration
Deep — wake-up receiver and retention only ~108 mW (over 99% of it self-discharge) ~4.1 years
Plus passive spatial sensing ~127 mW ~3.5 years
Plus full-CSI spatial sensing ~309 mW ~1.4 years

Deep-sleep duration does not depend on pack size. Energy and drain scale together, so a 4 kWh pack and a 20 kWh pack both reach ~4.1 years at 2%/month. A larger pack buys sleep duration only once there is a fixed load beside self-discharge: with full-CSI sensing running, the reference 3.9 kWh pack gives ~1.4 years where a 10 kWh pack would give ~2.4.

Two design consequences.

(1) Optimising the always-on wake-up circuit below ~10 mW is wasted effort — it disappears beneath self-discharge. The only term worth optimising is full-rate spatial sensing, because it alone is the same order as the chemistry.

(2) What bounds sleep is not the standby electronics — it is the cell chemistry. A deep-sleeping body runs its pack down in a few years even if the always-on wake-up circuit drew nothing at all.

The bound disappears at levels 3 and 4, where the dock both carries the body and offsets self-discharge.

3.7 Environmental harvesting — add-on

Peak solar irradiance is ~1000 W/m². The effective collecting area of a ~1.65 m body is ~0.45–0.6 m²; good cell efficiency is 20–25% — giving a maximum harvest of ~90–150 W.

Against that, a humanoid in motion draws hundreds to over 1000 W. Solar extends standby; it does not run a working body. It counts toward waiting time and toward charging at rest, never toward the operating budget.

3.8 Hard constraints

No configuration yields unlimited energy in free-running mode. This is battery physics and it is not negotiable — it is only widened by the six measures and by charging infrastructure.

"Unlimited" is reached only at levels 3 and 4, on external power, where the ceiling is the external supply rather than the pack.

This is the honest substitute for an inexhaustible source: not an infinite supply, but a finite budget plus an infrastructure that makes it sufficient.

Operating endurance: ⟦CTRL⟧. No figure is set here. The designer supplies the range and the trade; the controller picks the point — a large pack for long outings, a light one for short work, more or less sleep as context demands — within the ceiling derived in 02.3.