Week 10 — Grasping and Manipulation Day 5 of 7
Why This Matters
A robot arm is only as useful as the object it can hold. Engineers routinely spend six figures on a 6-axis arm, tune its motion planner, and then bolt on a $40 hobby gripper that drops parts on the factory floor. The end-effector is the last 5% of the kinematic chain but it causes a disproportionate share of failures.
Consider a real failure class. A pick-and-place cell uses a two-finger parallel gripper with 20 N of grip force to move 2 kg aluminum housings. The friction coefficient between the rubber pad and the anodized aluminum is 0.4. The safety factor looks fine on paper until the cell accelerates at 15 m/s² during a high-speed transfer. The required grip force jumps, the part slips 3 mm inside the jaws, and the downstream vision system rejects every unit. The arm was perfect. The gripper selection was wrong.
There is a second, quieter cost. Wrong end-effectors force engineers to add fixtures, vacuum cups, and custom fingers. Each addition adds mass at the wrist, which lowers the payload rating, which forces a bigger arm, which raises cost. Getting gripper selection right on day one avoids this spiral. This lesson gives you the numbers to make that decision before you buy hardware.
Core Concepts
The Four Gripper Families and When Each Wins
Most industrial grasping falls into four families. Each has a distinct force, stroke, and compliance profile.
| Family | Typical Grip Force | Stroke | Best For | Weakness |
|---|---|---|---|---|
| Two-finger parallel | 10–200 N | 5–100 mm | Rigid parts, high volume | No centering on irregular shapes |
| Three-finger centric | 20–150 N | 10–80 mm | Cylindrical parts, self-centering | Bulky, heavier at wrist |
| Vacuum (cup) | 0.1–0.6 bar vacuum | N/A (flat) | Sheet metal, glass, cardboard | Fails on porous or curved surfaces |
| Soft/Fin-Ray | 1–20 N | 20–150 mm | Fragile, irregular produce | Low repeatability, slow |
The decision rule is mechanical, not aesthetic. If the part is rigid and has two opposed parallel faces, use a two-finger parallel gripper. If it is cylindrical and must be centered, use three fingers. If it is flat, non-porous, and lightweight, vacuum wins on cost and cycle time. If it deforms under 20 N, go soft.
A concrete number: a Schunk MPG-plus 50 two-finger gripper delivers 215 N of grip force per finger at 6 bar with a 6 mm stroke. A Festo HGPL-14 delivers 110 N with a 14 mm stroke. Same footprint class, half the force, double the stroke. The choice depends entirely on part tolerance, not on the gripper’s marketing sheet.
Grip Force Budget: The Slip Calculation
Grip force is not a single number you copy from a datasheet. It is a budget you compute from the worst-case load. The dominant failure mode is slip during acceleration.
For a two-finger gripper holding a part by friction on two faces, the maximum tangential force before slip is:
$$F_{tangential,max} = \mu \cdot 2 \cdot F_{grip}$$
where $\mu$ is the friction coefficient and $F_{grip}$ is the force each finger applies. The factor of 2 appears because both fingers contribute normal force. If the robot accelerates the part at $a$, the inertial force is $m \cdot a$, and you need:
$$2 \mu F_{grip} \geq S \cdot m \sqrt{a^2 + g^2}$$
Here $S$ is a safety factor (use 1.5 to 2.0 for industrial cells), $m$ is part mass, $g = 9.81$ m/s², and the square root combines horizontal acceleration and gravity into a single resultant load. This combined-load form is the one most engineers forget. They size for gravity only and the part slips during the first fast move.
Worked example: $m = 2$ kg, $\mu = 0.4$, $a = 15$ m/s², $S = 1.5$.
Resultant acceleration: $\sqrt{15^2 + 9.81^2} = \sqrt{225 + 96.2} = 17.9$ m/s².
Required grip force per finger: $F_{grip} \geq \frac{1.5 \cdot 2 \cdot 17.9}{2 \cdot 0.4} = \frac{53.7}{0.8} = 67$ N.
The 20 N gripper fails by a factor of 3.4. This is exactly the failure described earlier. The fix is either a higher-force gripper, a higher-friction pad (polyurethane at $\mu = 0.9$ drops the requirement to 30 N), or a slower motion profile.
Mathematical Foundation
Gripper selection is a constrained selection problem. You want the cheapest, lightest gripper that satisfies force, stroke, and payload constraints simultaneously.
Force constraint. From the slip budget above:
$$F_{grip,req} = \frac{S \cdot m \sqrt{a_{max}^2 + g^2}}{2\mu}$$
Symbols: $F_{grip,req}$ is required per-finger force (N), $S$ is safety factor (dimensionless), $m$ is part mass (kg), $a_{max}$ is peak tool-center-point acceleration (m/s²), $g$ is gravitational acceleration (9.81 m/s²), $\mu$ is static friction coefficient between pad and part.
Stroke constraint. The gripper must open wider than the part plus positioning error:
$$s_{req} = w_{part,max} + 2\delta_{pos} + c$$
where $w_{part,max}$ is the largest part width (mm), $\delta_{pos}$ is the one-sigma positioning error of the vision system or feeder (mm), and $c$ is a clearance margin (typically 2–5 mm).
Wrist payload constraint. The gripper mass plus part mass must stay under the arm’s rated payload at the worst-case reach:
$$m_{gripper} + m_{part} \leq m_{payload,rated}(r)$$
where $m_{payload,rated}(r)$ is the payload rating at reach $r$ (mm from shoulder), which drops as reach increases. A UR10e is rated 12.5 kg at full reach but 15 kg near the base. A Fanuc M-10iD/12 is rated 12 kg across a narrower envelope.
Worked numeric example. Part: 1.5 kg cylindrical steel shaft, diameter 40 mm, tolerance ±1 mm. Vision error $\delta_{pos} = 0.5$ mm. Peak acceleration 10 m/s². Pad friction $\mu = 0.5$. Safety factor 1.8.
Force: $F_{grip,req} = \frac{1.8 \cdot 1.5 \cdot \sqrt{10^2 + 9.81^2}}{2 \cdot 0.5} = \frac{1.8 \cdot 1.5 \cdot 14.0}{1.0} = 37.8$ N.
Stroke: $s_{req} = 40 + 2(0.5) + 3 = 44$ mm.
Payload: gripper mass must be $\leq 12 - 1.5 = 10.5$ kg at the working reach.
A three-finger centric gripper like the Robotiq 3-Finger Adaptive delivers 70 N with 65 mm stroke at 2.3 kg mass. It clears all three constraints with margin. A cheaper two-finger unit with 40 mm stroke would fail the stroke constraint at 44 mm. The math picks the gripper, not the catalog order.
Practical Example
Scenario: a packaging cell moves 400 g cardboard boxes with a 300 mm × 200 mm top face. The robot is a UR5e (5 kg payload, 850 mm reach). Cycle time target is 1.2 s per pick. Vision gives ±2 mm position error.
Step 1 — pick the family. Cardboard is flat, porous, and lightweight. Vacuum is tempting but porous cardboard leaks air and reduces suction. A soft gripper is too slow for 1.2 s cycles. A two-finger parallel gripper with wide fingers gripping the box sides is the robust choice. Box side is 300 mm, so stroke must be at least $300 + 2(2) + 4 = 308$ mm. That exceeds most parallel grippers. Reconsider: grip the 200 mm dimension instead. Stroke needed: $200 + 4 + 4 = 208$ mm. Still large. Use a wide-stroke parallel gripper such as the Robotiq Hand-E (50 mm stroke) — no. Use a Schunk PGN-plus-P 200 with a 120 mm stroke plus custom fingers that reach around the box. Custom fingers add 0.8 kg.
Step 2 — force budget. Box mass 0.4 kg, $\mu = 0.6$ (rubber on cardboard), $a_{max} = 8$ m/s², $S = 1.6$.
$F_{grip,req} = \frac{1.6 \cdot 0.4 \cdot \sqrt{8^2 + 9.81^2}}{2 \cdot 0.6} = \frac{0.64 \cdot 12.66}{1.2} = 6.75$ N per finger.
Very low. Almost any gripper clears this. The binding constraint is stroke and payload, not force.
Step 3 — payload check. Hand-E mass 0.9 kg, custom fingers 0.8 kg, box 0.4 kg. Total 2.1 kg. UR5e rated 5 kg, so fine. But acceleration derating: at 850 mm reach with 8 m/s² acceleration, effective payload drops roughly 20%. Still 4 kg available. Passes.
Step 4 — cycle time. Hand-E closes in 0.2 s at full stroke. Vision and approach take 0.7 s. Total 0.9 s. Meets 1.2 s target.
The final selection: Robotiq Hand-E with custom 3D-printed fingers, rubber pads, 6.75 N grip force, 0.9 kg total wrist mass. The force math was nearly irrelevant. The stroke and cycle time drove the decision. This is common — most gripper mistakes come from sizing force when stroke was the real limit.
Common Pitfalls
1. Sizing grip force for gravity only. Engineers compute $F = mg$ and stop. The part then slips during the first acceleration. Why: inertial force adds vectorially to gravity, and at 15 m/s² the resultant is 1.8× gravity. Detect it by logging gripper slip sensors or by watching for parts that shift 1–3 mm in the jaws after fast moves. Fix by using the combined resultant formula above.
2. Ignoring friction pad degradation. A new rubber pad has $\mu = 0.9$. After 50,000 cycles with oily parts, it drops to 0.3. The gripper that worked in the lab drops parts in month three. Why: friction depends on surface condition, not just material. Detect it with periodic pull tests or by tracking drop rate over time. Fix by specifying pads with a 2× friction margin and scheduling replacement.
3. Undersizing stroke for positioning error. A gripper with 20 mm stroke gripping a 18 mm part seems fine until the vision system is off by 1.5 mm and the finger hits the part edge. Why: stroke must cover part size plus twice the positioning error plus clearance. Detect it by measuring grip failures correlated with part position. Fix by using $s_{req} = w + 2\delta + c$.
4. Forgetting wrist mass in payload derating. Adding a 2.5 kg gripper to a 5 kg arm leaves 2.5 kg for the part. Then acceleration derating cuts that to 2 kg. Then a 2.2 kg part drops the arm into overload. Why: payload ratings are static and measured at low acceleration. Detect it by checking the arm’s payload-vs-reach curve at your actual acceleration. Fix by including gripper and finger mass in the budget from day one.
5. Using vacuum on porous or textured surfaces. A vacuum cup rated for 5 kg on glass delivers 0.3 kg on cardboard because air leaks through the material. Why: vacuum force is $F = \Delta P \cdot A$, and leakage collapses $\Delta P$. Detect it with a vacuum switch that logs pressure during each pick. Fix by using foam-sealed cups or switching to mechanical gripping for porous parts.
6. Overlooking compliance for insertion tasks. A rigid two-finger gripper cannot insert a peg into a hole with 0.1 mm clearance without force sensing or a remote center compliance (RCC) device. Why: rigid grippers transmit all misalignment as lateral force. Detect it by measuring insertion force spikes. Fix by adding an RCC wrist or a compliant gripper such as the Robotiq Adaptive Gripper.
Key Takeaways
- Size grip force from the combined resultant of acceleration and gravity, not gravity alone. Use $F_{grip} \geq \frac{S \cdot m \sqrt{a^2 + g^2}}{2\mu}$ with $S = 1.5$–$2.0$.
- Stroke is often the binding constraint, not force. Compute $s_{req} = w_{part} + 2\delta_{pos} + c$ before comparing datasheets.
- Friction coefficient degrades with contamination. Specify pads with a 2× friction margin and plan replacement intervals.
- Include gripper and finger mass in the wrist payload budget, and derate for acceleration using the arm’s payload-vs-reach curve.
- Match gripper family to part geometry: parallel for rigid opposed faces, three-finger for cylinders, vacuum for flat non-porous, soft for fragile or irregular.
Tomorrow
Tomorrow we build a Python grasp simulation — you will model a two-finger gripper, compute grasp quality with the Ferrari-Canny metric, and test slip conditions against real part masses in code.