BoidKit

1 · Why fluid power

An electromechanical joint cannot separate where energy is converted from where force appears; a fluid one can. Conversion hardware — heavy, and worth building well exactly once — stays in the core. Pressure travels. Force turns up at the joint that asked for it, produced by something that weighs very little. The marginal axis is then valve + lines + bundle + geometry + sensor, against a motor, a reduction stage, bearings and a housing.

This does not make hydraulics automatically cheap or efficient. Seals, leakage, pressure losses, contamination, thermal management, valve precision, pump efficiency and fatigue all remain real engineering constraints. The claim is architectural: shared pressure makes a different cost and mass allocation possible. Whether that allocation survives realistic cycle counts is unresolved.

2 · Working fluid

Engineering target

Design work centres on water or a water/glycol mixture rather than petroleum hydraulic oil.

3 · Pressure and flow

Engineering target — not a validated lifetime rating

Pump architecture

Two pump cassettes so ordinary operation can be shared and some faults isolated. The chain runs:

4 · Artificial muscle

Active research

Joint torque comes from bundles of linear contractile fluid actuators. The exact construction can evolve; the architectural properties that matter are low actuator mass, low marginal material cost, inherent compliance, easy scaling by fibre count, parallelisation of load, and routing that resembles tendon rather than a motor housing.

Sizing one actuator for the whole joint load is the expensive path. An axis instead runs 4–10 fibres per side, several hundred across a body, and each is designed for its share — one tenth of the aggregate, before load-sharing, transient, fatigue and safety factors go on top. The clause doing the work there is approximately evenly. Uneven sharing is the failure mode that would take the economics with it.

Antagonistic joint

A joint is driven by opposing bundles analogous to flexor and extensor groups. One proportional 4/3 valve meters pressure and return between the two sides.

The valve uses progressive port geometry: small initial openings give fine flow resolution, further rotation opens a larger area for fast motion. The intent is to get fine and coarse behaviour from one cheap valve rather than duplicating metering hardware. Major load-bearing axes may carry passive load-hold elements so loss of hose pressure does not immediately collapse the body.

5 · Manifolds and control electronics

Prototype

Shared pressure and return galleries, valve seats, sensor ports and mounting collapse into one digitally fabricated unibody, parameterised from the morphology description rather than redrawn each revision. Early programme work included exactly this: a parameterised 3D-printable unibody manifold intended to replace many separate hydraulic transmission components, plus custom driver electronics and simulation of the muscle and valve system.

Zone controller responsibilities

• Encoder sampling
• Local pressure, temperature and contact channels
• Valve zero and calibration maps
• Setpoint interpolation
• Position, velocity and force limiting
• Local fault detection
• Deterministic command timestamps
• Graceful behaviour during upstream packet delay

Main compute should never have to micromanage valve PWM.

6 · Fabrication

Prototype — this is how the existing parts were made

Make the geometrically expensive parts; buy the commoditised ones.

A practical workshop is closer to a small digital-fabrication lab than a machine shop: FDM printers, a CNC router, a laser cutter, drill press and reaming tools, lapping fixtures for valve and sealing surfaces, heat sealing, winding and braiding fixtures, a pressure proof and burst fixture, flow and leak rigs, and ordinary electrical test equipment. Metal machining is still required for some shafts, wear surfaces, fittings and fixtures. The goal is not ideological avoidance of metal — it is to stop precision metal cylinders and custom geartrains from dominating the marginal cost of every joint.

7 · Qualification

Self-fabricated fluid-power components pass disciplined gates before they go near a robot. Home fabrication does not mean skipping pressure-vessel engineering.

8 · Using it from outside

Engineering target

BoidKit was conceived as a modular, plug-and-play actuation platform and stays one even as our own machines get more vertically integrated. Likely integration surfaces:

• A stable typed control API
• ROS / ROS 2 bridges where they reduce integration cost
• Recorded and replayable state and command streams
• Explicit mechanical, electrical and fluid interface definitions per module
• Simulator adapters generated from the same morphology description the real robot uses

9 · Open questions for this layer

1. What muscle geometry maximises lifetime-adjusted force per dollar rather than peak force alone?
2. Can water-hydraulic seals and valves stay low-leakage over useful cycle counts with mostly polymer components?
3. What pressure minimises total system mass after pump, line, valve, actuator and structural scaling are included?
4. How much accumulator volume is worth carrying?
5. Can flow be allocated predictively across many simultaneously moving joints?
6. What failure modes dominate after 10⁵–10⁷ cycles?
7. Can a progressive rotary valve provide both fine low-flow controllability and high peak flow from cheap servo torque?
8. What fabrication and finishing process produces acceptable leakage and deadband repeatably?
9. How should valve friction, fluid force and temperature drift be modelled locally?

Next — the machines →