Grippers & End-Effectors — Robotics Reference

Grippers & End-Effectors

See also (Tier 3 family index): End-Effectors Zoo

1. At a glance

The end-effector is the device bolted to the flange at the tip of a manipulator ([[Robotics/manipulator-design]]) — the part of the robot that actually touches the world. Everything upstream (kinematics, dynamics, control, trajectory generation) exists to put this device at a commanded pose with a commanded wrench; everything downstream (the part, the workpiece, the human) depends on whether it can hold, place, twist, weld, spray, or feel correctly. Pick the wrong end-effector and a perfectly engineered arm fails the application.

End-effectors split into two top-level classes:

• Grippers — devices that grasp a discrete object so the arm can transport, position, or insert it. Five dominant families: (a) parallel-jaw two-finger, (b) angular / multi-finger, (c) suction (vacuum), (d) magnetic, (e) soft / compliant. Plus emerging multi-finger dexterous hands (Shadow, Allegro, SVH, PSYONIC) for in-hand manipulation, and jamming grippers for irregular-part picking.
• Process tools — devices that act on a workpiece without lifting it. Spot-weld guns, MIG/TIG torches, spray nozzles, sanders, deburrers, drillers, glue dispensers, dispensing heads, screwdrivers.

In practice the same arm typically alternates between several end-effectors via a tool changer (ATI, SCHUNK SWS, Stäubli MPS, Destaco TC) so one robot serves multiple processes in a cell.

Where it sits in the stack: end-effector sizing is downstream of part geometry, payload mass, cycle time, and ROI, and upstream of grasp planning ([[Robotics/computer-vision-robotics]]), tactile feedback ([[Robotics/sensors-force-tactile]]), and compliant control ([[Robotics/impedance-control]]). The chosen end-effector also constrains arm payload: a 1 kg gripper + 2 kg part on a UR5e leaves only 2 kg headroom of the 5 kg rated payload.

First ask before selecting:

1. Part geometry + material. Rigid box flat surface → suction is almost always cheapest. Cylindrical / irregular → parallel-jaw. Ferromagnetic sheet → electro-permanent magnet. Soft / fragile / variable → soft gripper or compliance-augmented jaw.
2. Cycle time. Pneumatic two-finger close in ~50 ms; electric servo in ~200 ms; vacuum-on (with surface contact) in 30–80 ms; vacuum-off (with blow-off) in 20–40 ms. A 0.3 s/cycle line cannot afford a 200 ms gripper.
3. Payload + dynamics. Static grip force must hold weight; dynamic must hold weight × (1 + a/g) accounting for arm acceleration. Industrial rule of thumb: 30–50 % de-rate on catalogue grip force for production duty.
4. Human collaboration. Cobot end-effectors must respect ISO/TS 15066 power-and-force limits; pinch geometry, padded jaws, and rounded edges become design constraints, not afterthoughts.
5. Sensing needs. Position feedback (object detected?), force feedback (correct grip force?), slip detection ([[Robotics/sensors-force-tactile]]), vision integration (where to grasp?).
6. Flange + bus. ISO 9409-1 50 / 80 / 100 / 125 mm flange standardises the mechanical interface; cobots add M12 power + RS-485 / EtherCAT / Modbus RTU for the gripper bus.

2. First principles

Grasp analysis — form-closure vs force-closure

A grasp is form-closure if the contact geometry alone immobilises the object: the object cannot move infinitesimally in any direction without violating a contact constraint, regardless of friction. Three-finger envelope grasps and full caging grasps are form-closure; a simple parallel-jaw grasp on a sphere is not.

A grasp is force-closure if the contact wrenches the gripper can apply — within their friction cones — span the full 6-D wrench space. Friction matters here: the gripper can resist any external wrench by selecting an interior point of the friction-cone polytope. Practical 2-finger grasps on rigid parts rely on force-closure; form-closure is a luxury reserved for multi-finger and caging designs.

Antipodal grasp condition for two-finger:

• Two contact points $p_1,p_2$ with inward normals $n_1,n_2$.
• The line $\overline{p_1{p}_2}$ must lie inside the friction cone at both contacts: $\angle (\overline{p_1{p}_2},n_1)\le \arctan \mu ,\angle (\overline{p_2{p}_1},n_2)\le \arctan \mu$.
• Centre of mass must project onto the segment $\overline{p_1{p}_2}$ (avoids tipping moment).

Reference: Bicchi & Kumar 2000; Murray-Li-Sastry 1994 chapter 5.

Friction cone and grip force

Coulomb friction at a single point contact: tangential force is bounded by $|F_t|\le \mu F_n$. The set of admissible contact wrenches forms a cone of half-angle $\arctan \mu$ about the inward normal. Required grip force to hold a part of mass $m$ under gravity (vertical) with two opposing flat jaws and friction coefficient $\mu$:

$F_{\text{grip}}\ge \frac{mg}{2\mu }\cdot k_{\text{safety}}$

with safety factor $k_{\text{safety}}=2$ for steady lift, 3 for moderate acceleration, 5 for shock / unpredictable handling. The factor 2 in the denominator reflects two friction interfaces (one per jaw); a three-finger envelope grasp on a cylinder may have $k_{\text{friction}}=3\mu$ in the denominator.

Common $\mu$ between jaw material and part:

Jaw material	Part material	$\mu$
Aluminium (bare)	Aluminium	0.3
Steel (bare)	Steel	0.25
Polyurethane 90A	Aluminium	0.7
Polyurethane 90A	Glass	0.9
Nitrile rubber (NBR)	Cardboard	0.6
Silicone (Shore 30A)	Cardboard / paper	0.8
Gecko adhesive (Stanford SDM)	Glass	0.6–1.2 (preload-dependent)
Bare jaw	Oily / machined steel	0.05–0.10

Wrench space

The wrench $\mathbf{w}=(F_x,F_y,F_z,{\tau }_x,{\tau }_y,{\tau }_z{)}^T$ a gripper can exert on its object is constrained by contact friction cones and joint torque limits. Mapping joint torques $\mathbf{\tau }$ to wrench:

$\mathbf{w}=\mathbf{G}{\mathbf{f}}_c,{\mathbf{f}}_c\in {\mathcal{F}}_1\times {\mathcal{F}}_2\times \dots$

where $\mathbf{G}$ is the grasp matrix and ${\mathcal{F}}_i$ is the friction cone at contact $i$. The convex hull of representable $\mathbf{w}$ is the wrench polytope; large polytopes mean the gripper resists arbitrary disturbances, small polytopes mean specific axes are weak.

Suction (vacuum) physics

Vacuum gripping lifts via atmospheric pressure pushing on a sealed surface. Maximum holding force on a perfect seal:

$F_{\text{max}}=(P_{\text{atm}}-P_{\text{vac}})\cdot A_{\text{seal}}$

with $P_{\text{atm}}=101.3$ kPa, $P_{\text{vac}}$ the absolute pressure inside the cup (industrial systems achieve 20–30 kPa abs, i.e. 70–80 % vacuum), and $A_{\text{seal}}$ the effective sealed area (≈ 70 % of the cup’s nominal area for bellows; 90 % for flat). De-rate by 50 % for shear/tilt loading and by another 50 % for porous or curved surfaces.

Vacuum generation has two paths:

• Pneumatic Venturi ejector (Schmalz SCPSi, Festo OVEM, Piab piCOMPACT): compressed air at 0.4–0.6 MPa drives a converging-diverging nozzle, entraining cup air. Cheap, no moving parts, but inefficient (~70 % of compressed-air energy is dissipated). Fastest evacuation (30–80 ms for typical cup volume).
• Electric vacuum pump (Schmalz ECBPi/ECBPmini, OnRobot VG10): brushless DC diaphragm or rotary-vane pump driven directly off the cobot’s 24 V bus. Energy-efficient, no compressed-air infrastructure, slower (100–300 ms evacuation), louder at idle.

Magnetic gripping

Force on a ferromagnetic plate of cross-section $A$ in a perpendicular magnetic field $B$ at the plate-gripper interface:

$F=\frac{B^{2}A}{2{\mu }_0}$

with ${\mu }_0=4\pi \times 10^{-7}$ H/m. A 0.5 T field across a 1000 mm² pole face gives $F=0.25\cdot 10^{-3}/(2\cdot 1.257\cdot 10^{-6})\approx 99$ N. Real magnetic grippers achieve 50–80 % of theoretical because of air gaps, surface roughness, and non-saturation of the work piece.

Three magnet types:

• Permanent magnet — always on; release requires mechanical actuation (lift-off lever, cam). Used in low-volume material handling.
• Electromagnet — current-controlled, holds only when energised. Fails open on power loss (safety concern: holds object only while powered).
• Electro-permanent (EPM) — pair of magnets (AlNiCo + NdFeB) whose net field is switched by a brief current pulse through a coil. Holds with zero standby current; pulse to engage, pulse to release. Magswitch, Walmag, Sertek, Goudsmit are the major suppliers; EPM is the dominant choice for industrial magnetic grippers.

Soft / compliant grasping

A soft gripper deforms to match the object surface, distributing contact pressure rather than applying point loads. Two mechanisms:

• Pneumatic soft fingers (Soft Robotics Inc mGrip / FANUC FSC, Festo BionicSoftHand). A multi-chambered elastomer bladder inflates asymmetrically, curling around the object. Pressure 30–80 kPa; per-finger tip force 5–25 N. Inherent compliance handles unknown geometry; grip force is set by inflation pressure.
• Jamming granular gripper (Brown et al. 2010 PNAS, Empire Robotics — defunct; Univ. of Chicago / Cornell / MIT research). A latex membrane filled with ground coffee or glass beads is pressed onto the object; vacuum-evacuating the membrane jams the granules into a rigid form-fitted shell. Holding force 10–40 N typical; works on almost any geometry but slow (200–500 ms cycle).

Soft grippers trade speed and precision for adaptability. Cycle times 200–500 ms are common vs 50–200 ms for hard parallel-jaw.

Dexterous (multi-finger) manipulation

A multi-finger hand with 3–5 fingers and 10–25 DOF can execute in-hand manipulation: rolling, sliding, finger-gaiting, pivoting, regrasping the object without releasing it. The mechanics are a contact-rich hybrid system: each finger’s contact may stick, slip, or roll, and the controller must reason about contact states.

Reference designs: Salisbury 1985 (Stanford/JPL hand, 3-finger 9-DOF), Cutkosky 1989 (taxonomy of human grasps), Shadow Dexterous Hand E3M5 (20-DOF tendon-driven), Wonik Allegro (4-finger 16-DOF), Schunk SVH (5-finger 9-DOF), PSYONIC Ability Hand (6-DOF prosthetic).

Tool exchange

An automatic tool changer (ATC) is a two-part coupling: master plate on the arm flange, tool plate on each end-effector. Pneumatic locking (taper-cone + ball-detent) clamps the tool to the master in ~0.2 s; integrated pass-throughs carry compressed air, electrical signals, vacuum, and fluid lines. Standardised flange bolt-pattern per ISO 9409-1; vendors include ATI (MC-, QC-series), SCHUNK (SWS), Stäubli (MPS), Destaco (TC). Repeatability after re-coupling is typically ±0.005–0.025 mm depending on master size — adequate for most pick-and-place, sometimes marginal for precision insertion (force control covers the residual error).

3. Practical math + worked examples

Example A — Parallel-jaw sizing for a 200 g aluminium block

Spec: part 200 g aluminium, 50 × 30 × 20 mm. Grip on the 30 mm faces with polyurethane-coated jaws ($\mu =0.7$). Arm peak acceleration 5 m/s² (≈ 0.5 g) during pick-place.

Step 1 — total dynamic load. Worst-case (acceleration aligned with gravity, downward in pick): $a_{\text{eff}}=g+a_{\text{robot}}=9.81+5=14.81$ m/s². Weight under acceleration: $W_{\text{dyn}}=0.2\cdot 14.81=2.96$ N.

Step 2 — friction-cone grip force. Two opposing flat jaws:

$F_{\text{grip}}\ge \frac{W_{\text{dyn}}}{2\mu }\cdot k_{\text{safety}}=\frac{2.96}{2\cdot 0.7}\cdot 3=6.34\text{ N}$

Step 3 — select. Robotiq 2F-85 (max grip force 235 N, min 20 N, stroke 0–85 mm, mass 925 g): hugely over-spec but matches the UR cobot ecosystem with native ROS 2 driver + URCap. OnRobot 2FG7 (max 140 N, 70 mm stroke, mass 670 g) is the lighter alternative when payload headroom is tight on a UR5e. For a pneumatic line, SCHUNK MPG-plus 25 (160 N at 0.6 MPa, mass 130 g) is the fastest and cheapest if pneumatic infrastructure exists.

Step 4 — check pinch hazard. At 6.4 N grip force the cobot is far under the ISO/TS 15066 hand-pinch quasi-static limit (140 N back-of-hand quasi-static); the Robotiq 2F-85 also has integrated current-based pinch detection that backs off below 50 N if it senses unexpected contact during closure.

Example B — Suction cup for 4 kg cardboard box

Spec: box 4 kg, top surface 300 × 400 mm (free to grip anywhere on top), corrugated cardboard, lift-only motion (no tilt during transport).

Step 1 — required vertical force. $W=4\cdot 9.81=39.2$ N; dynamic factor 1.5 for robot acceleration → $F_{\text{req}}=59$ N.

Step 2 — single bellows cup option. Schmalz SAB 60×2.5 NBR-AS (60 mm Ø bellows, NBR rubber, anti-static). Catalogue holding force at 60 % vacuum (40 kPa abs): 140 N vertical / 70 N horizontal. Effective sealed area for a 60 mm bellows ≈ π·(0.024)² = 1.81·10⁻³ m². At 60 % vacuum: $F=(101.3-40)\cdot 10^3\cdot 1.81\cdot 10^{-3}=111$ N theoretical → 140 N rated assumes higher vacuum / larger effective area; for cardboard de-rate by 30 % for surface porosity → ~98 N effective. Comfortably > 59 N.

Step 3 — system selection. OnRobot VG10 with one 60 mm cup at one of two channels: simplest for a UR cobot deployment, electric pump on-board. Schmalz CobotPump ECBPMi with single port works equally well. If cycle-time critical: add a second cup → twice holding force, faster vacuum break (smaller volume per channel).

Step 4 — pump sizing. Schmalz ECBPMi delivers 12 L/min free-air at 25 kPa abs. Cup volume ≈ 50 cm³ = 0.05 L; evacuation time from atmosphere: $t\approx V\cdot \ln (P_{\text{atm}}/P_{\text{vac}})/Q=0.05\cdot \ln (4)/12\approx 6\times 10^{-3}$ min = 0.36 s — within typical cobot pick cadence.

Example C — Electro-permanent magnet for 10 kg steel plate

Spec: plate 10 kg, 200 × 200 × 3 mm, mild steel (low-carbon, ferromagnetic, ${\mu }_r>1000$).

Step 1 — required force. $W=98.1$ N; safety 2× for shock = 196 N.

Step 2 — Magswitch AR40. AR40 EPM module, 40 mm pole face, 130 N holding force at full contact on 6 mm plate, 90 N on 3 mm plate (de-rated for thinner work piece because of flux leakage). Single AR40 on 3 mm plate: 90 N — under spec.

Step 3 — alternatives. Two AR40 in parallel: 180 N (just under 196 N target — marginal). Single AR70 (70 mm pole, 230 N nominal on 6 mm, ~160 N on 3 mm): under spec. Magswitch AR50: 195 N on 6 mm, ~135 N on 3 mm. Best option: two AR50 in parallel (270 N on 3 mm) or one AR100 (520 N on 6 mm, ~365 N on 3 mm).

Step 4 — partial contact derate. If the gripper lifts from a plate edge where only 60 % of the pole face contacts: force drops 60–70 % (flux path is not closed). The system must specify full pole contact via mechanical centring or vision-guided positioning.

Step 5 — release verification. EPMs hold residual magnetism even when commanded off (~5–10 % of holding force). Verify drop via gripper-mounted Hall sensor (Honeywell SS49E, AKM AK8975) reading field at the pole face — if it doesn’t drop below threshold within 50 ms of the release pulse, fault.

Example D — Three-finger envelope grasp on a cylindrical workpiece

Spec: aluminium cylinder 80 mm Ø × 200 mm, mass 2.7 kg, must be lifted and rotated 90° about the horizontal axis (centrifugal load during slew).

Step 1 — required holding wrench. Static weight 2.7 · 9.81 = 26.5 N vertical. Rotation at 1 rad/s sustained with cylinder COG 0.1 m from rotation axis: centrifugal 2.7 · 0.1 · 1² = 0.27 N (negligible). Angular acceleration 5 rad/s² gives tangential 2.7 · 0.1 · 5 = 1.35 N. Dominant load is static gravity, with a torque component about the grasp centre of 26.5 · 0.05 = 1.32 N·m if the centre-of-mass falls 50 mm outside the grasp midplane.

Step 2 — three-finger force-closure on cylinder. With three contact points spaced 120° around the cylinder and friction μ = 0.6 (urethane-coated jaw on aluminium), the friction cones at the three contacts intersect over a full 6-D wrench polytope (Salisbury 1985). Per-finger normal force to resist gravity:

$F_n\ge \frac{mg}{3\mu }\cdot k_{\text{safety}}=\frac{26.5}{3\cdot 0.6}\cdot 3=44\text{ N}$

Step 3 — select. SCHUNK PZN+ 100 (3-finger centric gripper, 1300 N total grip force, stroke per finger 6 mm, mass 1.7 kg) — vastly over-spec but the standard industrial 3-finger centric for ~80 mm parts. For a cobot deployment, OnRobot 3FG15 (3-finger, force-controlled to 240 N total, stroke 35 mm per finger) is preferable; cobot-rated edges, programmable force per finger.

Step 4 — finger geometry. Custom V-groove or arc fingertips matched to the 80 mm diameter eliminate point contact, distribute force across a contact patch of 5–15 mm² per finger, and lift the effective μ by 10–20 % through micro-mechanical interlock.

4. Design heuristics

Selection by part type

• Flat rigid box, top accessible → suction (single or multi-cup). Fastest cycle, cheapest, scales easily. Default for case-packing, palletising, parcel handling.
• Cylindrical / prismatic rigid part → parallel-jaw with form-fitting fingertips. The fingertips are application-specific (V-grooves, custom CNC inserts, urethane-coated profiles).
• Soft / fragile / variable geometry (produce, baked goods, fabric) → soft pneumatic gripper. Soft Robotics mGrip (FANUC FSC), Festo BionicSoftHand, OnRobot Gecko, Yale OpenHand T42.
• Ferromagnetic sheet (automotive panels, appliance shells, sheet-metal stamping) → EPM magnetic gripper. Zero standby power; safe failure mode (holds on power loss).
• Very small parts (chip handling, watch components) → vacuum micro-nozzle or precision parallel-jaw with custom tips. SCHUNK MPG 16 / Festo HGRT.
• Mixed bin (unknown part) → jamming gripper (Empire pattern), multi-finger hand, or compliance + vision-guided parallel-jaw. The current state of the art for pure heterogeneous bin-picking is still vacuum + vision with re-grasping retries.

Cycle-time impact

Gripper open/close time is part of the cycle, not free overhead. Typical:

Gripper class	Open-close (ms)	Force settle (ms)
Pneumatic parallel-jaw (small, e.g. SCHUNK MPG 25)	30–60	10–20
Pneumatic parallel-jaw (large, e.g. SCHUNK PGN+ 100)	80–150	30
Servo parallel-jaw cobot (Robotiq 2F-85, OnRobot 2FG7)	150–400	50–100
Vacuum cup, evacuation (ejector)	30–80	—
Vacuum cup, evacuation (electric pump, small cup)	100–300	—
Vacuum release (blow-off)	20–40	—
EPM switch on/off	20–50	10
Soft pneumatic gripper inflation	150–400	50
Jamming gripper evacuation + lift	200–600	50

For a 0.3 s pick-place cycle (target on a delta-line packaging cell), gripper actuation cannot consume more than ~80 ms; that rules out servo parallel-jaw cobot grippers entirely, leaves vacuum ejectors or pneumatic miniature parallel-jaw.

Sensor integration on the end-effector

• Position feedback / part-present. Internal jaw-position encoder (every servo gripper has it); inductive or photoelectric sensor on the finger for binary “part present” (SICK MM, IFM IF6). Default for “did I pick something?” check.
• Force feedback at the jaw / fingertip. Built into Robotiq Hand-E (force-controlled jaw), OnRobot 2FG7, Weiss WSG-32 with FMSF fingertip. Reads 1–235 N at the jaw.
• Wrist-mounted 6-axis F/T sensor (ATI Axia-80, OnRobot HEX-E, Bota Rokubi 450) just behind the gripper. Reads contact wrench in any direction, decoupled from gripper internals. Required for force-controlled assembly (peg-in-hole, insertion).
• Tactile pads on the jaw face (XELA uSkin, Pressure Profile Systems DigiTacts, GelSight Mini camera-based). Reads pressure distribution + slip onset; used in grasp-quality and slip-detection control loops. See [[Robotics/sensors-force-tactile]].
• Vision in the gripper. Cognex In-Sight 2000, Keyence IV3, SICK Inspector — verify grasp location, orientation, or successful pick.

Compliance budget

The gripper plus its mounting must not be infinitely stiff: small misalignments (±0.5 mm typical between commanded TCP and actual part location) need to be absorbed somewhere or the part is dropped, crushed, or jammed.

• Passive compliance — Remote Center of Compliance (RCC) device (ATI RCC, SCHUNK FAS) between flange and gripper. A planar / angular spring stack that lets the gripper tip displace laterally by 1–5 mm and tilt 1–3° under modest force. Required for high-speed peg-in-hole on rigid arms.
• Active compliance — wrist F/T sensor + impedance / admittance control ([[Robotics/impedance-control]]). The arm itself yields under contact force; no mechanical compliance device. Standard on cobots with joint-torque sensing (Franka, iiwa) and on industrial arms equipped with an ATI Axia or Bota Rokubi sensor.
• Hybrid — passive RCC + active F/T; the RCC handles fast transients, the F/T loop corrects steady-state.

Payload de-rate

Gripper catalogue figures (max payload, max grip force) are usually for vertical static lift with the jaw axis horizontal. Real applications de-rate:

Condition	De-rate factor
Steady horizontal motion, low accel	0.7
1 g (9.8 m/s²) acceleration	0.5
Tilted grasp (jaw axis 30° off horizontal)	0.6
Slippery / oily surface	0.3
Porous surface (cardboard, foam)	0.4
Wet surface	0.4
Long-stroke gripper at extended jaw	0.7

For example a SCHUNK PGN+ 100 rated 750 N at the jaw drops to ~375 N under 1 g acceleration on a slightly oily steel part — apply the rule of thumb early to avoid in-deployment surprises.

Mechanical interfaces

• ISO 9409-1:2004 — robot mechanical interface, flange types A50 / A80 / A100 / A125 (50 / 80 / 100 / 125 mm bolt pattern). All cobots and industrial arms expose one of these flanges; all end-effectors mount via an adapter to the chosen flange. The flange has a precision pilot diameter (h7) that locates the tool concentric with the wrist axis.
• Power + comms (cobot) — M8 / M12 connector at the flange carrying 24 V, dGND, RS-485 or EtherCAT, and (often) 4 × digital I/O. UR’s tool flange has a single M8 8-pin connector exposing dual 24 V + dual digital I/O + serial. Franka uses an M8 for power and a separate M12 for high-speed bus.
• Pneumatic — for industrial arms, 4 mm or 6 mm push-fit fittings on the wrist; air routed inside the arm via the dressing kit.

Quick-change and tool exchange

For multi-process cells (e.g. weld + handle + dispense on one arm), an ATC (ATI MC-50 / QC-X10, SCHUNK SWS, Stäubli MPS) is mandatory. Sizing rule: pick the ATC by continuous moment capacity, not payload — moment is what fails first under dynamic load. ATI MC-50 handles 50 kg payload and 105 N·m continuous moment; MC-150 handles 150 kg / 425 N·m. Tool-change cycle 1–3 s including air-pin verification.

Cobot safety constraints

Under ISO/TS 15066 ([[Robotics/manipulator-design]]), the end-effector must not pinch, shear, or impact above the body-region thresholds. Practical end-effector design rules:

• Finger tips rounded ≥ 5 mm radius; no sharp 90° edges.
• Pinch points padded with > 5 mm of compliant material (urethane, NBR foam).
• Closing force limited at ≤ 110 N total for typical hand presence (the limit is body-region-specific; back-of-hand quasi-static is 140 N).
• Closing speed limited so kinetic energy at the jaw face is < 0.5 J (avoids transient impact violation).
• Integrated contact / force detection that backs off the close at < 50 % of nominal grip force when unexpected contact appears (Robotiq 2F-85 implements this in firmware via motor-current monitoring).

Maintenance and life

Class	Wear item	Typical MTBF / replacement	Note
Pneumatic parallel-jaw	Seals + air filter	5–10 × 10⁶ cycles	Replace seal kit at scheduled PM
Servo parallel-jaw	Ball-screw or rack-pinion	10⁷ cycles	Bearings then ball-screw wear
Vacuum cup	Rubber lip / bellows	6–18 months on cardboard; 3–6 yr on smooth steel	Wear depends on surface abrasion
Vacuum pump (diaphragm)	Diaphragm	~5000 hr	Ejector has no wear but uses compressed air
EPM	None significant	> 10⁸ switch cycles	Hall sensor verification recommended
Soft pneumatic	Elastomer (UV/ozone)	6–24 months	Protect from sunlight + ozone (welding cells)
Tool changer	Locking balls + air seal	10⁶–10⁷ cycles	ATI rates QC-X10 at > 5 × 10⁶

5. Components & sourcing

Industrial parallel-jaw

• SCHUNK — PGN+ / PGN-plus (universal, 2–4 finger, pneumatic), MPG / MPG+ (small, pneumatic), JGZ-S (long-stroke), JGP-S, PHL (parallel long-stroke). Workhorse industrial line.
• SMC, Festo — MHZ2 series (SMC), HGP / HGPP / HGPT (Festo). Equivalent pneumatic parallel-jaw lines, often cheaper, less catalogue accessory support.
• Destaco RobotiQ, Yamaha YHRG, PHD GRH — secondary suppliers.

Cobot parallel-jaw

• Robotiq — 2F-85 (235 N, 85 mm stroke), 2F-140 (125 N, 140 mm stroke), Hand-E (185 N, 50 mm stroke, force-controlled). URCap + Modbus RTU + native ROS 2 driver. The de-facto standard on UR cobots; supported on Doosan, Franka, KUKA iiwa via adapter.
• OnRobot — 2FG7 (140 N, 70 mm stroke), RG2 (40 N, 110 mm stroke), RG6 (120 N, 160 mm stroke), 2FG14 (140 N, 140 mm stroke), 3FG15 (3-finger). OnRobot Compute Box bridges to UR / Doosan / Yaskawa / Techman.
• SCHUNK — EGP (servo, small), EGN, EGL, EGP-C (collaborative-certified). Generally heavier than Robotiq / OnRobot but more rigid; Co-act EGP-C is the cobot-rated line with rounded edges.
• Festo — HGP-…-2 (cobot-ready).
• Weiss Robotics — WSG-32 / WSG-50 with FMSF fingertip force-sensing module (true 1-axis force feedback at the jaw).

Pneumatic 2 / 3-finger angular and centric

• SCHUNK — PZN+ / PZN-plus (3-finger centric), DPZ+ (2-finger centric), JGZ (long-stroke 3-finger), MPZ (small 3-finger).
• Destaco GA-series, PHD Grippers GRT/GRP, Bimba Manufacturing.

Vacuum systems

• Schmalz — CobotPump ECBPi / ECBPi mini / ECBPMi (electric, cobot-rated), Compact Ejector SCPSi / SCP / SCPS (Venturi), JumboFlex (heavy-duty industrial vacuum lifter), SAOL / SAX / SAB / SAOG cup catalogue, FQE/FQB family (filtration).
• OnRobot — VG10 / VGC10 (electric, 4-channel), VG10X.
• Festo — OVEL / OVEM (Venturi), OGVM (medium vacuum), VAS cup catalogue.
• Piab — piCOMPACT (Venturi), piCOBOT (cobot pump), BX / BL / FCF cup catalogue, Kenos KCS (3D-printed-tool format).
• ANVER, FIPA — cup and lifter specialists.

Magnetic grippers (electro-permanent)

• Magswitch — AR / QR series (AR20, AR40, AR50, AR70, AR100; small to medium), MagJig (low-cost workshop). Industrial standard for automotive panel handling.
• Walmag — ELM electro-permanent for steel handling.
• Sertek — automated EPM cells.
• Goudsmit Magnetics — ULM Pneumatic and ULM electro-permanent.

Soft / compliant grippers

• Soft Robotics Inc — mGrip (formerly the company; assets / product line acquired by FANUC in 2024 under FANUC FSC). Pneumatic soft fingers for produce, frozen food, baked goods.
• Festo BionicSoftHand 2.0 — pneumatic biomimetic hand (research / showcase).
• OnRobot Gecko — gecko-inspired dry adhesive on a flat pad; lifts smooth flat parts (glass, sheet metal) without vacuum or magnet. 6–10 N/cm² preload-dependent.
• Yale OpenHand T42, Model O, Model T — open-source designs.
• Empire Robotics (defunct ~2016) — Versaball jamming gripper; design pattern still active in academia (Brown 2010 PNAS).

Dexterous robotic hands

• Shadow Robot Dexterous Hand E3M5 — 20-DOF (24-DOF total inc. wrist), tendon-driven, biomechanically inspired. Used in research and OpenAI’s Rubik’s-cube demo.
• Wonik Robotics Allegro Hand — 16-DOF 4-finger research hand. Direct-drive joints, no tendons.
• SCHUNK SVH — 9-DOF 5-finger anthropomorphic hand. Industrial-supported.
• PSYONIC Ability Hand — 6-DOF prosthetic-grade hand (also available for humanoid / research integration). Medicare-funded for amputees; published in IEEE J-BHI (Akhtar 2023).
• Sanctuary AI Phoenix hand — 21-DOF humanoid hand (proprietary).
• DLR Hand-II / Hand Arm System, IIT iCub hand, KIST KITECH hand — academic / research.

Tool changers

• ATI Industrial Automation — QC-X10 (5 kg), MC-15 (15 kg), MC-50, MC-100, MC-150, MC-440, MC-1100 (1100 kg). Pneumatic locking + electrical + pneumatic + fluid pass-throughs.
• SCHUNK SWS series (SWS-005 to SWS-510).
• Stäubli MPS series.
• Destaco TC series.

F/T-integrated EOAT

• Robotiq FT 300-S wrist sensor + Hand-E gripper.
• OnRobot HEX-E / HEX-H wrist sensor + 2FG7 / RG2 / RG6 gripper.
• Bota Systems Rokubi 450 + adapter to any gripper.
• ATI Axia80 + adapter.

Specialty

• Process tools — Genesis, Cemsa, Centerline (spot-weld); Fronius, ESAB, Lincoln (MIG/TIG torches); 3M, Festool (sanders); SCREWMAT, Atlas Copco MicroTorque (screwdriver heads); Nordson, Graco (dispensing); FANUC IRPickTool (vision-integrated).

6. Reference data

Gripper family × payload × cycle × cost

Family	Payload range	Cycle (open-close)	Per-unit cost (USD)	Standby power
Pneumatic 2-finger (small, MPG class)	< 0.5 kg	30–80 ms	$300–$800	none (air on close)
Pneumatic 2-finger (medium-large, PGN+)	0.5–10 kg	80–200 ms	$700–$3000	none
Servo 2-finger cobot (Robotiq 2F-85, OnRobot 2FG7)	0.5–5 kg	150–400 ms	$4500–$8000	< 5 W
Pneumatic 3-finger angular (PZN+)	0.5–10 kg	100–250 ms	$1200–$4000	none
Vacuum (single cup + ejector)	0.2–4 kg	30–80 ms evac	$400–$1500 (incl. ejector)	compressed air only on
Vacuum (electric pump, multi-cup)	0.5–10 kg	100–300 ms	$3000–$8000	~30 W on
EPM magnetic (AR40-class)	1–15 kg	20–50 ms switch	$1500–$6000	none (pulse only)
Soft pneumatic 3–4 finger	0.1–2 kg	150–400 ms	$5000–$15 000	air on close
Multi-finger dexterous (SVH, Allegro)	0.5–5 kg	200–1000 ms	$30000–$150 000	10–50 W
Process tool (weld gun, spray)	n/a (no part lift)	task-specific	$2000–$50 000	task-specific
Tool changer (ATI MC-50)	50 kg	1–3 s exchange	$4000–$8000	air during change

Cobot flange compatibility (common end-effector mounts)

Cobot	Flange ISO 9409-1	Native bus to gripper	Electrical at flange
Universal Robots e-series	A50 (50 mm)	RS-485 over M8 8-pin	24 V dual + 4 × DI/DO
Franka FR3	A50 (50 mm)	Internal Modbus / Franka Hand bus	24 V + dGND + RS-485
KUKA iiwa 7/14	A50 (50 mm)	Media flange (varies: electric, pneumatic, IO)	24 V + EtherCAT (option)
Doosan A/M/H	A50	RS-485 + 24 V via M8	dual DI/DO
Techman TM	A50 + integrated camera mount	RS-485 + 24 V	DI/DO
Kassow K-series	A50	RS-485	dual DI/DO
Fanuc CRX cobot	A50 (with adapter)	RS-485 / EtherCAT	DI/DO
ABB GoFa	A50	EtherCAT	DI/DO
Kinova Gen3	Custom flange (Kinova-specific) → A50 adapter	RS-485	24 V

Vacuum cup material × surface × recommended vacuum

Cup material	Surface	Recommended vacuum	Cup geometry
NBR (nitrile)	Smooth metal, glass	60–80 % (20–40 kPa abs)	Flat or 1.5-bellows
Silicone (FDA grade)	Food, soft, hot (< 200 °C)	60–80 %	Flat or 1.5-bellows
Polyurethane (PU)	Sharp-edge sheet metal, oily steel	50–70 %	Flat
EPDM	Outdoor, ozone-exposed	60–80 %	Flat or bellows
Foam (open-cell)	Porous (cardboard, fabric)	30–50 % (lower vacuum, higher flow)	Foam pad
HNBR	Hot stamping, > 150 °C	60–80 %	Bellows
Anti-static (NBR-AS / SI-AS)	Electronics, PCB	60–80 %	Flat with anti-stat path

Grip force ranges by class

Class	Force range	Typical models
Micro precision parallel	1–20 N	SCHUNK MPG 16, Festo HGRT
Small pneumatic parallel	20–200 N	SCHUNK MPG 25 / 40, SMC MHZ2-16
Medium pneumatic parallel	200–1000 N	SCHUNK PGN+ 50 / 80, Festo HGPT-25
Large pneumatic parallel	1000–5000 N	SCHUNK PGN+ 200 / 300, Destaco GR50
Servo parallel cobot	5–250 N	Robotiq 2F-85 (5–235), OnRobot 2FG7 (3–140)
3-finger pneumatic	100–2000 N total	SCHUNK PZN+ 64 / 100 / 160
Vacuum, single cup 60 mm	up to ~140 N at 60 %	Schmalz SAB 60, Piab BX 65
Vacuum, single cup 100 mm	up to ~350 N at 60 %	Schmalz SAB 100, Piab BX 110
EPM, AR40-class	90–130 N (de-rated for thin work)	Magswitch AR40
EPM, AR100-class	365–650 N	Magswitch AR100
Soft pneumatic per finger	5–25 N	Soft Robotics mGrip, Festo BionicSoftHand

ISO 9409-1 flange standard sizes (mechanical interface)

Flange	Bolt circle Ø (mm)	Pilot Ø (mm)	Bolt count × thread	Used on
A31.5	31.5	25 h7	4 × M5	Mini arms (Yaskawa MotoMini, ABB IRB 120)
A40	40	31.5 h7	4 × M5	Small industrial / mini cobot
A50	50	40 h7	4 × M6	Cobot standard (UR, Franka, KUKA iiwa, Doosan, Techman)
A63	63	50 h7	6 × M6	Medium industrial
A80	80	60 h7	6 × M8	Medium-large industrial
A100	100	80 h7	6 × M10	Large industrial (KR50, IRB 4600)
A125	125	100 h7	8 × M10	Heavy industrial (KR210, M-2000)
A160	160	125 h7	8 × M12	Super-heavy (KR1000, M-2000)
A200	200	165 h7	12 × M12	KR1000 Titan, M-2000iA/2300

7. Failure modes & debugging

• Slipped part during fast motion. Symptom: part shifts in jaw mid-trajectory or drops. Cause: grip force below dynamic load. Fix: (a) raise grip force (servo gripper) within payload de-rate, (b) reduce arm acceleration, (c) add tactile slip sensor ([[Robotics/sensors-force-tactile]]) and close grip-force loop on slip detection, (d) switch to higher-μ jaw material (urethane → silicone).
• Dropped suction part from leaky surface. Symptom: vacuum sensor reads pressure rising above setpoint mid-lift. Cause: porous cardboard, perforated metal, dust on seal. Fix: (a) switch to bellows cup that conforms better, (b) higher-flow vacuum pump to outpace the leak, (c) larger cup or multi-cup with check valves so a single leak doesn’t sink the manifold, (d) anti-static cup if static is pulling dust into seal.
• Crushed fragile part. Symptom: part deformed / cracked after grip cycle. Cause: rigid gripper at max force on a soft part. Fix: (a) use force-controlled gripper with the lowest viable target (Hand-E, 2FG7), (b) add silicone or foam jaw pad (5–10 mm Shore 30A), (c) use soft pneumatic gripper at low pressure (30–40 kPa).
• Pneumatic leak at fittings. Symptom: rising compressor duty cycle, audible hiss, slow gripper close. Fix: leak-check with soapy water; replace push-fit seal or O-ring; verify air-line pressure regulator stays at 0.5–0.6 MPa under demand.
• Electric gripper stall under high inertia. Symptom: motor over-current fault during close on a heavy or jammed part. Fix: reduce close speed (most servo grippers have configurable speed); pick a heavier-duty model; preload the jaw position before final close to reduce travel.
• Soft gripper UV / ozone degradation. Symptom: elastomer fingers brittle, micro-cracks visible, holding force dropped. Cause: sunlight, welding-cell ozone, gamma sterilisation. Fix: opaque shroud during storage; HNBR or EPDM for ozone exposure; scheduled replacement every 6–24 months.
• Magnetic residue holding part after release. Symptom: EPM commanded off but part still stuck for 0.5–2 s. Cause: residual flux in EPM and work piece. Fix: (a) pulse de-magnetisation cycle (alternating short pulses of opposite polarity), (b) Hall sensor at pole face verifies field < threshold before release-confirm, (c) brief lift-with-shake motion.
• Vacuum-pump diaphragm wear. Symptom: gripper holds, but evacuation time creeps up over months. Cause: diaphragm fatigue at ~5 000 hr. Fix: replace pump head (Schmalz ECBPi has user-replaceable diaphragm kit), or switch to Venturi if compressed air is available.
• Tool-change misalignment. Symptom: after auto-change, gripper has 0.1–0.5 mm offset from previous calibration. Cause: ball-detent locking insufficient seating; air pressure low. Fix: verify air-pressure switch reads ≥ 0.5 MPa during change; clean taper-cone interface; replace ATC seals.
• Cable abrasion at the flange. Symptom: intermittent gripper comms drop or 24 V short. Cause: external cable harness fatiguing at joint-6 flex zone. Fix: route through hollow shaft / dressing kit; use chainflex cable; replace on schedule.
• F/T sensor drift post-impact. Symptom: wrist F/T reports phantom 5–10 N force when arm is at rest. Cause: thermal drift after motor heat-soak, or strain-gauge yield from prior over-load. Fix: (a) auto-tare on every cycle start, (b) calibration check (rotate sensor through gravity in known orientations and verify reading matches m·g of the gripper), (c) replace if drift > 5 % FS.
• Cobot ISO 15066 force-limit exceeded by gripper. Symptom: cobot E-stop on force-limit fault. Cause: gripper close speed or close force exceeds the body-region threshold during PFL. Fix: reduce close speed; reduce target force; add padded jaw face; if persistent, escalate to caged operation (ISO 10218 mode) for the affected motion.
• Wrong part in gripper. Symptom: arm proceeds with wrong-shaped part, downstream process fails. Cause: no part-present verification. Fix: integrate part-present sensor (inductive at jaw, vision on gripper); enforce verification before motion.
• Sticky / oily part slips even at full force. Symptom: μ collapses to 0.05–0.1; no realistic grip force holds. Fix: surface prep (degrease + air-knife), or anti-slip jaw coating (urethane sintered with carbide grit), or switch family (suction with high-vacuum + foam cup, or vision-guided regrasp at a clean spot).
• Vacuum cycle bottleneck on multi-cup manifold. Symptom: one cup is on a leaky part, manifold pressure cannot reach setpoint, pump stalls. Fix: per-cup check valves so non-sealing cups isolate themselves; segmented manifold with multi-channel pump (OnRobot VG10 with 4 independent channels).

8. Case studies

Robotiq 2F-85 — the cobot reference gripper

Robotiq launched the 2F-85 in 2013; over a decade later it remains the most-deployed cobot gripper globally with > 100 k units shipped (Robotiq). Design choices that won:

• Underactuated 4-bar finger linkage. Each finger has two phalanges driven by a single motor; the linkage enables both parallel pinch (rigid object) and encompassing power grasp (round object) without runtime control logic — geometry alone switches modes based on contact location.
• One motor, brushless DC + planetary, drives both fingers symmetrically through a worm gear (self-locking, holds without power).
• Grip force 5–235 N, programmable in 1 N increments; closing speed 20–150 mm/s programmable. Stroke 0–85 mm. Internal position encoder reports jaw width to ±0.1 mm.
• Native plug-and-play on Universal Robots via URCap; works on Franka FR3, Doosan A-series, KUKA iiwa via adapter and ROS 2 driver robotiq_2f_85_driver.
• Total mass 925 g including controller; 925 g of payload headroom lost on a UR5e (5 kg max), 925/16 ≈ 6 % on a UR16e — typically acceptable.

The 2F-85 sits at a sweet spot: enough force for 90 % of cobot pick-place tasks, cheap enough at ~$7 500 list to amortise across a single line, and integrated enough that an applications engineer can be productive in hours rather than days. It is not the right choice for ultra-fast cycles (the ~200 ms open-close kills throughput), heavy industrial parts (235 N is light), or precise insertion (Hand-E is better there because of its true force feedback). For the broad cobot middle of the market, it remains the default.

Amazon Robotics — the parallel-jaw to suction transition

Amazon Robotics’ warehouse picking systems have evolved through several end-effector generations:

• Kiva (acquired 2012) — floor robots moved entire shelves to human pickers. No robot end-effector.
• Sparrow (2022) — first product-picking arm. Used a hybrid suction-cup + simple parallel-jaw to handle ~65 % of inventory items. The vacuum cup grabs flat-top items (boxes, polybags); the parallel-jaw handled cylindrical / awkward items the cup could not seal on.
• Robin / Cardinal (2022–2024) — package sorting arms that switched to multi-cup vacuum gripper with 4–6 active channels and per-cup pressure feedback for grip-quality estimation.
• Sequoia / Digit (2023–2024) — integration of Agility Robotics Digit humanoid for tote handling, with custom under-actuated end-effectors that grasp tote-handles directly.

The architectural lesson: Amazon’s per-item value calculation drove them away from generic mechanical grippers toward family-specific end-effectors that handle the long tail of geometries through grip-quality inference and retry — not through richer hand mechanics. Vision + vacuum + retry beats a 20-DOF hand at warehouse-scale throughput as of 2025.

PSYONIC Ability Hand — prosthetic-grade dexterity at humanoid price

The PSYONIC Ability Hand (released 2021, peer-reviewed in Akhtar 2023 IEEE J-BHI) is a 6-DOF, 6-actuated-finger anthropomorphic hand designed initially as a prosthetic upper limb. Notable design choices:

• 6 motors (one per finger + a thumb-rotation motor) driving tendon-routed fingers; each finger has multiple joints driven through a 4-bar / tendon coupling.
• Touch-sensitive fingertips with embedded force sensors driving vibrotactile feedback (in prosthetic mode) and slip-grip control (in humanoid integration).
• Compliant fingers that bend out of the way under impact (no rigid fracture), survivable from drops and collisions where rigid hands break.
• Open API + ROS 2 integration; same hardware deployed on amputees (Medicare-funded) and on humanoid research platforms (Apptronik, several university platforms).

PSYONIC’s success demonstrates that prosthetic and humanoid end-effector requirements have converged: both need impact-survivable, sensor-instrumented, dexterous hands at $20–50 k unit cost. The Ability Hand’s life cycle proves out the design at scale (thousands of hours per amputee wearer) before it reaches a research humanoid — a maturity path no purely-research hand can match.

Schmalz / Robomotive — vacuum at industrial throughput

Schmalz’s vacuum tooling underpins the majority of automotive body-in-white panel handling, case-packing of consumer goods, and end-of-line palletisation. A representative deployment: a single ABB IRB 6700 with a Schmalz FXP-SVK matrix-style vacuum gripper (multiple zoned cups arranged in a programmable grid) lifts car-door inner panels off a stamping press at one part every 4 s, sustained 24/7. The gripper itself massed ~40 kg of the IRB 6700’s 200 kg payload — vacuum systems scale to the largest industrial arms because the vacuum pump and pneumatics can be mounted base-side, leaving only the cup matrix at the flange.

Architectural lessons: at industrial scale, vacuum is preferred not because it’s clever but because it’s boring, cheap, and well-understood. Cups are consumables ($20–80 each, replaced in seconds); the rest of the system has no moving parts at the end-effector. When the application admits a vacuum solution (rigid + flat-ish + non-porous), it dominates.

9. Cross-references

• manipulator-design — companion note; the arm that carries the end-effector.
• kinematics-dh — TCP transform from the last joint to the gripper centre.
• dynamics-rigid-body — end-effector mass + inertia enters the arm dynamics.
• motors-electric — servo motors inside electric grippers.
• sensors-force-tactile — wrist F/T, joint torque, jaw-mounted tactile pads, slip detection.
• impedance-control — compliant grasping and force-controlled insertion.
• trajectory-generation — pick-approach-grasp-depart trajectories that touch through the gripper.
• path-planning — grasp-pose planning and approach-direction sampling.
• computer-vision-robotics — grasp planning from RGB-D point clouds (Dex-Net, GraspNet, AnyGrasp).
• sensors-pose-motion — vision and 6D pose estimation for the part to grasp.
• materials-polymers — urethane, silicone, NBR, EPDM jaw and cup materials.
• fasteners-bolts — ISO 9409-1 flange bolts (M5–M12, grade 8.8 minimum).
• op-amps — strain-gauge bridge conditioning on force-feedback grippers.
• electric-motors — first-principles for servo gripper actuator selection.
• robotics-control — URScript / KRL / RAPID / KAREL gripper commands.

10. Citations

1. Bicchi, A. & Kumar, V. “Robotic Grasping and Contact: A Review.” Proc. IEEE ICRA, 2000, pp. 348–353. Survey of force-closure, form-closure, and grasp-quality metrics.
2. Murray, R.M., Li, Z. & Sastry, S.S. A Mathematical Introduction to Robotic Manipulation, CRC Press, 1994. ISBN 978-0-8493-7981-9. Chapter 5: multi-fingered hand kinematics and grasping.
3. Mason, M.T. & Salisbury, J.K. Robot Hands and the Mechanics of Manipulation, MIT Press, 1985. ISBN 978-0-262-13205-8. Foundational text on multi-finger grasp mechanics.
4. Salisbury, J.K. & Craig, J.J. “Articulated Hands: Force Control and Kinematic Issues.” International Journal of Robotics Research, 1(1):4–17, 1982.
5. Cutkosky, M.R. “On Grasp Choice, Grasp Models, and the Design of Hands for Manufacturing Tasks.” IEEE Transactions on Robotics and Automation, 5(3):269–279, 1989. The taxonomy of human grasps that guides anthropomorphic hand design.
6. Brown, E., Rodenberg, N., Amend, J., Mozeika, A., Steltz, E., Zakin, M., Lipson, H. & Jaeger, H.M. “Universal robotic gripper based on the jamming of granular material.” Proceedings of the National Academy of Sciences, 107(44):18809–18814, 2010. DOI:10.1073/pnas.1003250107. The jamming-gripper paper.
7. Galloway, K.C., Polygerinos, P., Walsh, C.J. & Wood, R.J. “Mechanically programmable bend radius for fiber-reinforced soft actuators.” IEEE ICRA, 2013, pp. 2189–2196. Soft pneumatic finger mechanics.
8. Akhtar, A., Aghasadeghi, N., Hargrove, L. & Bretl, T. “Mechatronic design and clinical performance of the Ability Hand multi-articulating myoelectric prosthesis.” IEEE Journal of Biomedical and Health Informatics, 27(8):3850–3861, 2023.
9. Yamaguchi, A. & Atkeson, C.G. “Combining Finger Vision and Optical Tactile Sensing: Reducing and Handling Errors While Cutting Vegetables.” IEEE-RAS Humanoids, 2016, pp. 1045–1051.
10. Ma, R.R. & Dollar, A.M. “On dexterity and dexterous manipulation.” Proc. IEEE ICAR, 2011, pp. 1–7.
11. Hammond, F.L., Mengüç, Y. & Wood, R.J. “Toward a modular soft sensor-embedded glove for human hand motion and tactile pressure measurement.” IEEE/RSJ IROS, 2014, pp. 4000–4007.
12. Pierce, B., Kemper, K. & Albu-Schäffer, A. “Robust adaptive cooperative manipulation.” Autonomous Robots, 36(1-2):85–98, 2014.
13. Mahler, J., Liang, J., Niyaz, S., Laskey, M., Doan, R., Liu, X., Ojea, J.A. & Goldberg, K. “Dex-Net 2.0: Deep Learning to Plan Robust Grasps with Synthetic Point Clouds and Analytic Grasp Metrics.” Robotics: Science and Systems, 2017. DOI:10.15607/RSS.2017.XIII.058.
14. Fang, H.-S., Wang, C., Gou, M. & Lu, C. “GraspNet-1Billion: A Large-Scale Benchmark for General Object Grasping.” IEEE/CVF CVPR, 2020, pp. 11444–11453.
15. SCHUNK Gripper Catalog 2024. SCHUNK GmbH & Co. KG, current edition. PGN+, MPG, EGP, EGN, SVH product lines.
16. Robotiq 2F-85 / 2F-140 / Hand-E Instruction Manual. Robotiq Inc., rev 2024.
17. OnRobot 2FG7 / RG2 / RG6 / VG10 / VGC10 Data Sheets. OnRobot A/S, 2024 catalogue.
18. Schmalz Vacuum Components & Systems Catalogue 2024. J. Schmalz GmbH; cup, ejector, ECBPi pump, FXP-SVK matrix gripper documentation.
19. Piab piCOMPACT / piCOBOT / Kenos / BX cup Catalogue. Piab AB, 2024.
20. Festo BionicSoftHand 2.0 Technical Brief. Festo SE & Co. KG, 2023.
21. Magswitch AR series Technical Specifications. Magswitch Technology Worldwide.
22. Soft Robotics Inc / FANUC mGrip Brochure. FANUC Corporation, 2024.
23. Shadow Robot Dexterous Hand E3M5 Technical Specification. The Shadow Robot Company.
24. Wonik Robotics Allegro Hand Technical Specification. Wonik Robotics Co.
25. ATI Industrial Automation — Tool Changer MC-, QC-X10 Manuals. ATI document numbers 9620-XX series, current revisions.
26. ISO 9409-1:2004 — Manipulating Industrial Robots — Mechanical interfaces — Part 1: Plates.
27. ISO/TS 15066:2025 — Robots and robotic devices — Collaborative robots. PFL force thresholds applied at the end-effector.
28. EN 1525:1997 — Safety of industrial trucks — Driverless trucks and their systems. Pinch / clamping force limits referenced for cobot grippers.
29. Stanford OpenHand Project / Yale OpenHand Documentation. Stanford and Yale GRAB Lab, open-source hardware.
30. Hayati, S. & Mirmirani, M. “Improving the absolute positioning accuracy of robot manipulators.” Journal of Robotic Systems, 2(4):397–413, 1985. Calibration model that closes flange-to-TCP accuracy on which gripper repeatability budgets depend.

---

Session log:

node ~/.claude/bin/obsidian-research.mjs log "Built Robotics/end-effectors.md Tier 1 deep note"