MODULE 2
From Prototype to Production

Motion You Can Repeat

Industrial robotics & factory automation — and why the demo cell is not the production line.
FramtidR — Advanced Manufacturing Technologies
Section 1

The Robot as Mechatronics

Loops inside loops: motors, gearboxes, encoders, and what makes motion repeatable.
FramtidRMotion You Can Repeat2 / 30
Mental model
A Robot Is Loops Inside Loops
A robot is not just a motor that turns — it is a stack of control loops running thousands of times a second.
Cell: PLC / MES / safetyKinematics & trajectoryJointloopcommands infeedback out
Commands flow inward; feedback flows back out. Lose any layer and the cell fails in a way the demo never showed.
FramtidRMotion You Can Repeat3 / 30
Inside one axis
What's Actually Inside One Axis
  • AC brushless servomotor. High torque for its size, switched by electronics, no brushes to wear out.
  • Precision reduction gearbox. Trades speed for torque while keeping play (backlash) very low.
  • High-resolution encoder. Optical or magnetic; reports angle and speed all the time.
  • Embedded servo drive. Runs the fast control loop right at the joint.
The gearbox raises torque and makes each encoder count worth a smaller step at the output.
motorencoderstrain-wave gearflangeembedded drive
Exploded view: motor → encoder → strain-wave gear → output flange, with a local drive closing the loop at the joint.
FramtidRMotion You Can Repeat4 / 30
Backlash, part one
Strain-Wave (Harmonic) Gears
  • Wave generator. Oval hub inside a thin ball bearing — this is the input.
  • Flexspline. Bendy toothed cup (the output); it has two fewer teeth than the ring.
  • Circular spline. Stiff ring with inside teeth, fixed to the housing.
  • Two teeth per turn. Each input turn moves the flexspline two teeth → high reduction, small size.
Near-zero play because teeth stay meshed and pre-loaded. "Backlash-free for the life of the gear" is a maker's claim.
wave gen.engageengage
The oval shape pushes the teeth into mesh at two opposite zones along the long axis.
FramtidRMotion You Can Repeat5 / 30
Backlash, part two
Harmonic vs Cycloidal Reducers
Rule of thumb: harmonic for small, light wrists; cycloidal for heavy base and shoulder joints.
Harmonic / Strain-wave
small, light wrists
  • Near-zero play (backlash)
  • Very high reduction per stage
  • Lower weight
  • Handles less shock
Cycloidal / RV
heavy base & shoulder axes
  • Low play (~1 arc-minute)
  • Load shared over many pins
  • Takes high shock; very stiff
  • Heavier, long life
FramtidRMotion You Can Repeat6 / 30
Servo loop
How a Joint Holds Its Position
  • Current / torque loop. Innermost, ~tens of kHz; sets winding current, which sets torque.
  • Velocity loop. ~kHz; asks for torque to hold a target speed.
  • Position loop. Outermost; asks for speed to drive position error to 0.
  • The encoder closes them all. Error = asked − measured; the drive trims current to shrink it.
A stiff, fast loop means the arm bends less under load and holds its position.
−Pos PIDVel PIDCur PIDMotor /Gear / Loadencoder feedbackouterfastest
Three nested PID loops; each outer loop sets the target for the inner one.
FramtidRMotion You Can Repeat7 / 30
ISO 9283
"Repeatable" Is Not "Accurate"
  • Repeatability. How tightly it comes back to the same taught pose, time after time.
  • Accuracy. How close it gets to a commanded pose in real-world space.
  • ISO 9283. Sets how to test pose accuracy, repeatability, drift and settling over ≥30 cycles in a test cube.
  • In practice. Repeats to ~0.02–0.2 mm, but real accuracy is often much worse — so offline cells need calibration.
repeatable,not accuraterepeatableand accurate
Tight cluster off-centre (left) vs tight cluster on the bullseye (right). The gap is why calibration exists.
FramtidRMotion You Can Repeat8 / 30
Section 2

The Math That Commands Motion

Transforms, forward vs inverse kinematics, the Jacobian, and the speeds that blow up.
FramtidRMotion You Can Repeat9 / 30
Spatial math
One Matrix for Where-and-Which-Way
  • Every frame has both. Position and direction in 3D — base, each link, tool, part.
  • 4×4 homogeneous transform. Rotation R (3×3) + shift p (3×1) + [0 0 0 1].
  • It belongs to SE(3). The set of rigid-body moves (turn plus shift).
  • Chain it to walk the arm: T₀ₙ = T₀₁ · T₁₂ · … · T₍ₙ₋₁₎ₙ.
R3×3p0 0 0 1zxyframe triad
Turn and shift packed into one object — "where is the tool relative to the base?" becomes matrix multiplication.
FramtidRMotion You Can Repeat10 / 30
FK vs IK
Telling an Arm Where to Reach
  • Forward kinematics (FK). Joint angles → tool pose. Direct, one answer, easy.
  • Inverse kinematics (IK). Wanted tool pose → joint angles. The hard but useful way.
  • Many possible answers. Elbow-up vs elbow-down, wrist flips; extra-axis (7+) arms have endless options.
  • Solvers choose by rules. Stay within joint limits, avoid self-collision, keep near the current pose.
FKjoint anglesone poseIKone poseelbow upelbow down
Two arm shapes reach the same point — the controller must pick which solution to use.
FramtidRMotion You Can Repeat11 / 30
Singularities
The Speeds That Blow Up
  • Jacobian J. Links joint speeds to tool speed: ẋ = J·q̇.
  • Flip it to plan. Wanted tool speed → needed joint speeds.
  • Singularity: det J → 0. J loses rank; the arm loses one direction it can move.
  • Paths can ask for ∞ speed. Planners watch J and slow down or steer around these spots.
DOF lostq̇ → ∞arm nearly straight (singular)det(J)0
At a singular pose the arm can't produce motion along one direction; det(J) dips to zero.
FramtidRMotion You Can Repeat12 / 30
Section 3

Mobility & Perception

Off the fixed stand: AMR vs AGV, SLAM, machine vision, and visual servoing.
FramtidRMotion You Can Repeat13 / 30
Mobile robotics
AMR vs AGV
  • AGV — Automated Guided Vehicle. Follows fixed guides: tape, wire, barcodes, reflectors.
  • A blocked path stops it. Predictable, but not flexible.
  • AMR — Autonomous Mobile Robot. Onboard LiDAR and cameras build a map and find a new route on the fly.
  • Trade-off. AGV: cheap, predictable, needs guides; AMR: flexible, sensor-rich, software-heavy.
halts on lineAGVreroutes aroundAMR
Same obstacle: the AGV stops on its printed line; the AMR's sensor cone plans a path around it.
FramtidRMotion You Can Repeat14 / 30
SLAM
Knowing Where You Are While Mapping It
SLAM builds a map of an unknown space while tracking where you are in it — each one needs the other.
LiDARweak in bare hallsWheel odometrydriftsIMUdrifts over timeFusion(e.g. EKF)Pose estimateMap
Different sensors are fused so each covers the others' weak spots — reliability comes from fusion, not one star sensor.
FramtidRMotion You Can Repeat15 / 30
Machine vision
From Light to a Decision
A vision system turns light into decisions through fixed steps — the result is only as good as the image going in.
1
Image capture
lens, sensor, lighting
▶
2
Clean-up
remove noise, boost contrast
▶
3
Find features
edges, blobs, key points
▶
4
Make sense of it
locate / measure / classify
2D = position & presence in a flat view; 3D = depth, pose, volume. Lighting is half the job — most failures are lighting, not the code.
FramtidRMotion You Can Repeat16 / 30
Visual servoing
Letting the Camera Steer the Arm
  • Closed vision-motion loop. Vision corrects the motion; the motion gives a new view.
  • Image-Based (IBVS). Acts on the error seen in the image; no full 3D pose needed.
  • Position-Based (PBVS). First work out the object's 3D pose, then move toward it.
  • Why it matters. Track and grab moving parts on a conveyor instead of needing perfect fixturing.
CameraControllerRobotimage erroractpart moves in viewconveyor part
Camera → image error → controller → robot → part moves in view → back to camera.
FramtidRMotion You Can Repeat17 / 30
Section 4

Building & Guarding the Cell

Atoms vs bits, collaborative safety, tooling, layout, and plant integration.
FramtidRMotion You Can Repeat18 / 30
Atoms vs bits
Same Word, Two Machines
Physical robots move atoms; RPA bots move bits.
Industrial robot
moves atoms
  • Physical work: weld, paint, assemble
  • Cost: machine, floor space, guarding
  • Fails: mechanical, calibration drift
  • Changes when: the physical part changes
RPA bot
moves bits
  • Software on screen: emails, invoices, DBs
  • Cost: licences, scripting; no hardware
  • Fails: breaks if the UI or access changes
  • Changes when: the screen changes
FramtidRMotion You Can Repeat19 / 30
Collaborative safety
Sharing the Floor With People, Safely
  • "Collaborative" = the whole application. Not the robot alone; the same arm can be safe when slow or deadly when fast.
  • ISO 10218 (Parts 1 & 2). Part 1 = the robot; Part 2 = the cell, system and integration.
  • ISO/TS 15066. Four methods: monitored stop, hand guiding, speed & separation, power & force limiting.
  • 2025 update. 15066 was merged into ISO 10218-2:2025; the field now says "collaborative applications."
fully guardedcollaborative appRPA bothigh speed/force → low human proximityFour collaborative methodssafety-rated monitored stophand guidingspeed & separation monitoringpower & force limiting
A spectrum of speed/force vs human proximity, with the four collaborative methods.
FramtidRMotion You Can Repeat20 / 30
EOAT
The Part That Touches the Work
The arm only positions; the end-of-arm tooling does the job — and it's where most cells succeed or fail.
▤
Mechanical jaw
2- or 3-finger; flexible, forgiving on part-to-part change
◯
Vacuum / suction
fast on flat sealed parts; fails on porous or warped ones
■
Magnetic
ferrous parts only, very fast; left-over magnetism & double-picks
⚙
Process tool
weld torch, nozzle, screwdriver — the gripper is the operation
FramtidRMotion You Can Repeat21 / 30
Cell layout
Drawing the Cell: Reach, Guarding, Flow
  • Reach & envelope. Check coverage with the EOAT fitted, not just the bare wrist.
  • Guarding. Fixed fences, light curtains, area scanners, interlocked gates — sized to the stopping distance.
  • Material flow. Parts go in and out without a person reaching into live motion.
  • Access. Service and teach access without switching off the safety system.
The integrator's risk assessment (per ISO 10218-2) drives guarding choices, not preference.
light curtain(stopping dist.)infeedoutfeedgate
Top-down plan: dashed reach circle, fixed fence, light curtain at the load opening, infeed/outfeed.
FramtidRMotion You Can Repeat22 / 30
Integration
The Robot Is One Node: PLC, MES & Recovery
  • PLC. Real-time, fixed-timing controller of conveyors, sensors, interlocks and start/stop signals.
  • MES. Sends out work orders, tracks each unit, records quality and history.
  • Handshakes. Digital I/O or a fieldbus tell the cell when a part is present, clamped, checked, cleared.
  • Error recovery is the dividing line. Detect, isolate, recover and log a fault — not just stop and wait.
Robot + sensorsPLC (real-time)MES (orders, quality)fault flowDetectIsolateRecoverLogResume
Three-tier stack with a one-fault recovery loop: a production cell expects faults, a prototype is surprised by them.
FramtidRMotion You Can Repeat23 / 30
Section 5

Prototype → Production

Payback, the integration valley, and when not to automate.
FramtidRMotion You Can Repeat24 / 30
ROI
Does It Pay? The ROI of a Cell
  • A money decision, not just engineering. Hardware is often the smaller half of the cost.
  • The hidden half. Integration, programming, safety sign-off, setup, maintenance and spares.
  • Returns. Less labour, more output, fewer defects/scrap, steady quality — but less flexibility.
  • Payback = cost ÷ yearly savings. Use is the lever: a 3-shift cell pays back far faster than a 1-shift one.
robotEOATintegrationguardingcertification3 sh.2 sh.1 sh.break-even by shifts
Installed cost stack and payback by utilization — illustrative teaching figures, not sourced benchmarks.
FramtidRMotion You Can Repeat25 / 30
The integration valley
A Demo Cell Is Not a Production Line
A demo works once; a line must work every shift, for years.
Cycle time vs takt time
the cell must finish a part faster than demand needs
Reliability / MTBF
tiny per-cycle failure rates add up to daily stops at volume
Safety certification
the guarded, risk-checked, ISO version is slower and costs more
PLC / MES integration
the robot is one node; it must hand off to conveyors and scanners
Calibration drift & recovery
heat, wear and crashes shift accuracy; detect, recover, log
FramtidRMotion You Can Repeat26 / 30
Maturity
When Not to Automate
Automation is a tool, not a goal. It's the wrong answer when:
  • ✓Volume is low or up-and-down. too few cycles to spread the cost over; payback never comes
  • ✓The product changes all the time. frequent redesigns happen faster than you can re-program and re-tool
  • ✓Parts vary far too much. heavy variation beats fixturing and vision; people handle the odd cases cheaply
  • ✓The process itself isn't steady. automating a broken process just makes bad parts faster
  • ✓Skill or judgment is key. fine assembly, handling exceptions, craft work where people still win
FramtidRMotion You Can Repeat27 / 30
Recap
Module Summary & What to Ask Your Vendor
  • Joints. Nested control loops; gearboxes (harmonic, cycloidal) buy precision by removing backlash.
  • Kinematics. Transforms, FK/IK, Jacobian, singularities — programmable, sometimes impossible.
  • Mobility & vision. AMR/AGV, SLAM, fusion, visual servoing free robots from fixed paths and fixtures.
  • The cell. EOAT, guarding/layout, PLC/MES integration, error recovery — where projects live or die.
  • Category & safety. Physical vs RPA vs cobot; respect ISO 10218 / 15066 → 10218-2:2025.
  • Production ≠ demo. Takt, MTBF, certification, drift, recovery — and it must pay back at real-world use.
Five questions for any vendor: repeatability at my payload and speed? Cycle time vs my takt? Independently checked MTBF? Which safety method and certification? How does it recover from a fault and talk to my MES?
FramtidRMotion You Can Repeat28 / 30
Further Reading
Robot hardware
  • Harmonic Drive — strain-wave gear principle — harmonicdrive.net/technology
  • Techman Automation — cycloidal / RV reducer specs
  • Robotiq — end-of-arm tooling types & selection
Kinematics
  • Clemson Open Textbooks — forward kinematics, 4×4 transforms, SE(3)
  • Columbia CS 4733 — Jacobians and kinematic singularities
  • MIT OpenCourseWare 2.12 — Introduction to Robotics — CC BY-NC-SA 4.0
Vision & mobile
  • OTTO / Rockwell — AGV vs AMR comparison
  • AI Review (Springer) — LiDAR–IMU–camera SLAM fusion survey
  • Scientific Reports — visual servoing (IBVS vs PBVS) tracking
Safety standards
  • ISO 10218-2:2025 — scope & absorption of ISO/TS 15066 — ISO OBP
  • ISO 9283 — robot pose accuracy & repeatability test methods
  • ISO 13855 — positioning of safeguards w.r.t. approach speeds
FramtidRMotion You Can Repeat29 / 30
⚠

Before you rely on this

This series was made by FramtidR with the help of AI. AI can make mistakes, including about people. Please check this information before you use it.

Made with AI help • checked by FramtidR
FramtidRMotion You Can Repeat30 / 30