Chapter 45: Automation and Robotics

Era span: 1788 governor → present · Difficulty: high
Requires: Ch 15, Ch 35, Ch 36, Ch 43
Unlocks: Ch 46
Data snapshot: volatile figures in this chapter (prices, capacities, deployment counts, regulation, and capability claims) reflect published sources through 2024 unless dated otherwise; check current data before planning.

Robotics is feedback control embodied: sense → decide → act, closed in a loop fast enough to matter. Watt's flyball governor (1788) was the first industrial robot in spirit; everything since adds sensing and computation to that loop.

Figure 45.1. The sensor return path is the defining difference between open- and closed-loop control. PID combines present, accumulated, and trend information; tuning is process-specific and requires bounded commissioning, not a universal recipe.

45.1 Control Theory Core

Feedback: measure output, compare to target, correct. Open-loop machines repeat blindly; closed-loop machines achieve targets despite disturbances.
PID controller (developed across early industrial control): proportional action responds to present error, integral action accumulates error and can remove offset, and derivative action responds to rate of change and can damp response. PID remains common, but no defensible universal share applies to every industrial loop.
Servomechanisms: motor + encoder + drive closing position/velocity loops at kHz rates — every robot joint is a servo.
Discrete-event control above continuous loops (state machines sequencing actions); safety-rated stops as separate hardwired paths (software never owns emergency stop).

Scope note: the classic Ziegler–Nichols ultimate-gain recipe can be aggressive and is not valid for every process. If used as an initial experiment, isolate the plant, impose safe bounds, record the response, calculate candidate gains, and verify them under a supervised commissioning procedure. Add derivative filtering, output limits, anti-windup, setpoint shaping, and process-specific knowledge. Never force oscillation in a hazardous or slow plant merely to fill in a card.

45.2 From NC to Industrial Robots

Numerical control (MIT, 1952): punched-tape machine tools machining to programmed coordinates — CNC made precision REPRODUCIBLE (Ch 20's tolerances democratized). G-code became manufacturing's Latin.
Unimate (1961): die-casting robot arm at GM — dangerous, repetitive job automated first (a recurring pattern: robots take the four D's — dirty, dull, dangerous, dear).
Robot anatomy: articulated arms (6 DOF = full pose), SCARA (fast planar assembly), delta pickers (lightning pick-and-place), cartesian gantries (heavy/precise). End effectors define capability: grippers, welders, sprayers, suction. Harmonic drives pack huge ratios into thin joints; encoders close the loops.

Arm	Moves	Wins at	Example
Articulated 6-DOF	Full pose anywhere	Welding, general	Auto body lines
SCARA	Fast planar + Z	Assembly	Electronics
Delta	Lightning pick	100+ picks/min	Food, pharma
Cartesian gantry	Heavy + precise	Big panels, CNC-like	Aerospace, machine tending
Cobot (force-limited)	Beside humans	Small shops, no cage	ISO/TS 15066 limits

45.3 Sensing

Machine vision: structured light/stereo depth, defect inspection at line speed — quality control becomes automatic statistics (Ch 47).
Force-torque sensors enable assembly by FEEL (peg insertion, compliance tasks); lidar/radar/ultrasonic give spatial awareness outdoors.
MEMS inertial units (Ch 35) put navigation-grade IMUs in every phone/drone — the quiet sensor revolution.

Vision buy-order: presence check (part there?) → measurement (in tolerance?) → guidance (where to grip?) → inspection (defect-free?). Light the scene before upgrading the camera — optics and illumination decide more than megapixels (Ch 19 returns here).

45.4 Mobile Robots and Drones

AGVs (fixed paths) → AMRs (self-navigating warehouse fleets; Kiva→Amazon scale: hundreds of thousands of units moving shelves to pickers — logistics throughput multiplied while human walking eliminated).
Drones: brushless motors + lithium batteries + MEMS IMUs + flight controllers = the hobby-to-industrial stack. Fixed-wing endurance vs multirotor agility split missions: survey/spraying/mapping/delivery niches each have a winner. Regulatory integration (airspace rules) lags capability everywhere — budget for governance friction.
Surgical robotics (teleoperation with tremor filtration, camera steadiness): precision augmentation where human judgment steers machine steadiness.

Platform	Endurance	Payload	Wins
Multirotor	20–40 min	Grams–kg	Inspection, mapping, camera
Fixed-wing	Hours	kg	Survey, spraying, corridors
VTOL hybrid	1–2 h	kg	Runway-free range
Ground AMR	Shifts	100s kg	Warehouses (shelves to picker)
Teleop surgical	Procedure-length	Instruments	Tremor-free precision

45.5 Humanoids and the Honesty Section

Figure 45.2. Payoff is highest where geometry is known: structured cells first, collaborative cells second, humanoids when hands catch up. Theater demos live on the right; warehouse economics lives on the left.

Humanoid general-purpose robots remain HARD: bipedal balance solved-ish, but dexterous manipulation (fingers adapting to novel objects) lacks its transistor moment. Theater demos ≠ warehouse economics. Pragmatic doctrine:

Automate STRUCTURED environments first (factories, warehouses — geometry known, parts standardized).
Deploy mobile manipulation where ROI clears without anthropomorphism.
Watch lab-to-field transfer of tactile sensing and learned grasping (Ch 46) as the humanoid-enabling frontier — but don't gate production plans on it.

45.6 Deployment Doctrine

Safety warning: industrial arms crush and trap — stored energy drives through power loss, fenceless cobots still strike heads and throats, and jam-clearing under power kills maintainers. Cage or force-limit per ISO/TS 15066, wire hardwired e-stops on a separate path, lock out before clearing jams, and commission with risk assessment plus exception-handling drills signed before go-live.

Collaborative robots can be designed for selected shared-workspace tasks, but “cobot” does not mean automatically safe without caging. Risk depends on payload, reach, speed, tooling, entrapment, impact, process hazards, maintenance access, and the applicable robot and machinery standards.
Maintenance robotics (pipe crawlers, blade-climbing rigs) extends asset life in hazardous inspection (Ch 30-adjacent safety gains).
Lights-out factories are mostly hybrid cells: automation handles repetition; humans handle exceptions — design workflows around exception HANDLING capacity.
Labor transition policy matters (Ch 47): automation's productivity dividend gets politically confiscated if retraining fails — plan retraining as part of deployment cost, not afterthought.

Capability gate: automation is an economic decision. Compare total installed and lifecycle cost with the task's required throughput, quality, uptime, supervision, safety, and integration—not labour cost alone. A robot cheaper than an hour of human work may still be a poor investment if downtime, tooling, safety, or exception handling dominates. Automation changes labour allocation; it does not automatically dissolve demographic constraints.

Cell commissioning list: risk assessment + hardwired e-stop (separate path!) → guarded vs force-limited decision (ISO/TS 15066 table) → exception handling plan (who clears jams?) → MTBF spares (grippers wear, cables flex-fail) → retraining roster signed before go-live.

45.7 The Robotics Papers

Watt's flyball governor (1788) prompted Maxwell's "On Governors" (1868) — control theory's founding paper written about steam engines; Minorsky's USS New Mexico automatic-steering trials (1922–23) introduced PID-like ship autopilots; Ziegler–Nichols tuning rules (1942) gave practitioners recipes. Control matured through ships and chemical plants before robots existed.
NC machining: Parsons Corporation's helicopter-blade templates + Air Force funding + MIT's servomechanisms lab demonstrated numerical control March 9, 1952; APT programming language standardized it — defense contracts buying manufacturing language again (Ch 49's armory pattern).
Unimate #001 entered GM's Ewing Township die-casting line January 1961. Devol filed "Programmed Article Transfer" in 1954; Engelberger — who met Devol at a cocktail party discussing Asimov's robot stories — built the business. Science fiction recruiting industrial capital, on record.
Shakey the Robot (SRI, 1966–72) needed pathfinding; Hart, Nilsson & Raphael invented A* (1968) to give it one — every GPS route and game AI since inherits that algorithm from a wheeled box with antennae.
Honda's P2 (1996) surprised the field with bipedal autonomy → ASIMO (2000); Boston Dynamics' DARPA-funded quadrupeds and Atlas stunts kept dynamics visible while warehouse economics quietly won adoption: Kiva Systems (founded 2003) sold to Amazon for $775 million (2012); Roomba (2002) put millions of robots in homes doing ONE task well.
da Vinci cleared by FDA July 2000 — teleoperative surgery's precision/tremor-filtering case study (§45.4).
Cobot safety formalized late: ISO/TS 15066 (2016) specified force/pressure limits by body region — collaborative robotics required a published injury-biomechanics standard before insurers and regulators allowed shared workspaces at scale.

45.8 Four-D Priority List

Die-casting, spray/paint, welding, palletising, and inspection are common early candidates because they combine hazardous, repetitive, or quality-critical work. Prioritise by measured task risk, variation, ergonomics, throughput, integration burden, and failure consequence. “Danger and tedium first” is a useful scan, not a universal priority rule or a substitute for worker participation.