The 5 Sensors That Give a Robot the Sense of Touch
Your hand has about 17,000 touch receptors. They tell you if something is hot, sharp, heavy, soft, wet, or vibrating — all at once, in real-time, without you thinking about it. A robot has none of that. So how do you give a machine the ability to feel? With five very specific sensors, each one replicating a different aspect of human touch.
When you pick up an egg, you don't think about how much force your fingers are applying. You don't calculate the coefficient of friction between your skin and the eggshell. You don't consciously adjust your grip based on the weight. Your hand just does it — thousands of nerve endings firing simultaneously, your brain processing the signals in milliseconds, your muscles adjusting in real-time so the egg doesn't slip and doesn't crack.
Now try programming a robot to do that. Without touch sensors, the robot has no idea how hard it's gripping. It might crush the egg. It might barely hold it and drop it. It can't tell the difference between an egg and a baseball. It can't feel the egg starting to slip. It's completely, utterly blind to the physical world in a way that's hard to appreciate until you try to build a machine that interacts with it.
That's why touch sensing is one of the hardest problems in robotics — and one of the most important. Every robot that interacts with the physical world needs some form of touch. Factory robots that assemble electronics. Surgical robots that operate on tissue. Prosthetic hands that help amputees feel again. And in our case, a robot arm that tattoos human skin at exactly the right depth, with exactly the right pressure, on surfaces that are soft, curved, and constantly moving as the person breathes.
Here are the five sensors that make it possible.
Sensor 1: Force/Torque Sensor — The Robot's Muscle Memory
Cost: $40 (DIY) to $3,000+ (industrial) · Size: 45mm disc · Weight: 50-200g
A force/torque (F/T) sensor is the most important touch sensor on any robot that physically interacts with the world. It measures two things: force (how hard something is pushing or pulling in three directions — left/right, forward/back, up/down) and torque (how much something is twisting around three axes). That's six measurements, updated thousands of times per second, giving the robot a complete picture of every physical interaction happening at its tool tip.
Think about what happens when you push a thumbtack into a corkboard. Your thumb feels the resistance increase as the pin penetrates the cork. If the cork is soft, there's barely any resistance. If you hit a dense spot, the resistance spikes. If the tack bends sideways, you feel a lateral force telling you the angle is wrong. A force/torque sensor gives a robot this exact same information.
The sensor itself is elegantly simple in concept. It's a precision-machined metal disc with thin, flexible sections (called "flexures") that bend microscopically when force is applied. Tiny strain gauges bonded to these flexures measure the bending. More bending means more force. By placing gauges at specific locations and angles, the sensor can decompose any applied force into its six components (Fx, Fy, Fz, Tx, Ty, Tz) simultaneously.
The industry standard is the ATI Mini45 — a 45mm diameter sensor that costs $3,000-5,000 new. It can measure forces as small as 0.05 Newtons (about the weight of a nickel) and updates at 7,000 Hz. For context, the force of a tattoo needle entering skin is about 0.5-3 Newtons, so a sensor with 0.05N resolution can detect changes of about 1/60th of the total force. That's like feeling the difference between pressing your finger on a table and pressing your finger on a table with a single sheet of paper on it.
The cheaper alternative: Stanford researchers published an open-source design called CoinFT that you can build yourself for about $40. It uses a 3D-printed flexure structure with six bonded strain gauges and HX711 amplifier boards. It's not as precise as the ATI, but it's good enough for prototyping.
Real-World Use:
In our tattoo robot, the F/T sensor is mounted between the robot arm's wrist and the tattoo needle assembly. It measures exactly how hard the needle is pressing into the skin. A software control loop running 1,000 times per second reads the force value and adjusts the arm's position to maintain a constant target force — which directly controls needle depth. If the skin gets thicker (moving from the inner wrist to the forearm), the resistance increases and the sensor detects it instantly, allowing the AI to increase push force to maintain the same 1.5mm penetration depth.
Sensor 2: Pressure-Sensitive Film — The Robot's Fingertips
Cost: $8-15 per sensor · Size: Paper-thin film, any shape · Weight: <1g
While a force/torque sensor tells you the total force at the wrist, pressure-sensitive films tell you exactly where the pressure is distributed across a surface — like the difference between knowing "something is pressing on your hand" versus knowing "something is pressing on the tip of your index finger, right side, with about 2 Newtons of force."
These sensors are called FSRs (Force-Sensitive Resistors). They're made from a thin polymer film with conductive particles embedded in it. When you press on the film, the particles get squeezed closer together and the electrical resistance drops. More pressure = lower resistance. An analog-to-digital converter reads this resistance change and translates it to a force value. The whole thing is thinner than a credit card, completely flexible, and can be cut to any shape with scissors.
The most common version is the FlexiForce A201, which measures 0 to 25 Newtons across a small circular pad about 10mm in diameter. It costs about $8 per sensor. You can stick it anywhere — on a robot's fingertip, on the inside of a gripper, on the bottom of a foot, or in our case, on the silicone fingertips of our skin-stretching mechanism.
The more advanced versions — like the BioTac sensor used in some prosthetic hands — have multiple sensing elements arranged in a grid under a soft silicone skin, giving a spatial map of pressure across the entire fingertip. These cost $5,000+ each and are primarily used in research. For most robotic applications, individual FSR pads at $8 each get the job done.
Real-World Use:
Our tattoo robot's V3 design has four mechanical fingers that physically grip and stretch skin during tattooing (like a human tattoo artist's free hand). Each fingertip has an FSR sensor bonded under a silicone pad. This lets the robot know exactly how hard each finger is gripping the skin. The AI limits grip force to 5 Newtons per finger — firm enough to stretch the skin taut, gentle enough not to bruise. If one finger detects the skin slipping (sudden drop in pressure), it signals the other fingers to compensate instantly.
Sensor 3: MEMS Accelerometer — The Robot's Vibration Sense
Cost: $3-5 per sensor · Size: 3mm × 5mm × 1mm · Weight: <0.1g
Put your finger on a running washing machine. You feel vibration — a rapid, oscillating force too fast for your brain to perceive as individual pushes. Your Pacinian corpuscles (deep pressure receptors in your skin) are specifically tuned to detect vibration between 40 and 800 Hz. This is why you can feel a phone vibrate in your pocket through layers of fabric, but you can't feel steady pressure through the same fabric — different receptors, different capabilities.
A MEMS (Micro-Electro-Mechanical Systems) accelerometer does the same thing for a robot. It's a microscopic silicon structure — literally smaller than a grain of sand — that moves when vibration occurs. The movement changes the capacitance between tiny silicon plates, which is measured electronically. Modern MEMS accelerometers like the ADXL345 measure vibration in all three axes simultaneously, at frequencies up to 3,200 Hz, with resolution fine enough to detect vibrations smaller than the width of a human hair.
These sensors cost about $3 each and are tiny — the entire chip is smaller than your pinky fingernail. They're the same sensors in your smartphone that detect when you rotate the screen or count your steps. For robotics, they're mounted directly on the tool or end effector and provide a vibration "fingerprint" of whatever the robot is doing.
Paired with a MEMS microphone (like the INMP441, $2 each), the accelerometer gives the robot both tactile vibration data and acoustic data. The microphone captures the sound of the vibration in the air, while the accelerometer captures the vibration through physical contact with the tool. Together, they provide a rich vibration signature that AI can analyze for anomalies.
Real-World Use:
A tattoo needle vibrating at 120 Hz through normal skin produces a specific vibration pattern. When the needle hits scar tissue, the vibration frequency shifts. When it hits cartilage (near the ear), the pattern changes dramatically. When a needle is wearing out and becoming dull, the vibration amplitude increases. Our AI learns these patterns and can distinguish "normal skin at correct depth" from "scar tissue — go slower" from "danger — cartilage or bone detected — stop immediately." The accelerometer is why the robot can "feel" what's under the skin without seeing it.
Sensor 4: Bioimpedance Sensor — The Robot's Tissue Scanner
Cost: $15-40 · Size: 20mm × 14mm PCB + electrodes · Weight: 5g
This is the one sensor on this list that gives a robot a sense humans don't naturally have. Bioimpedance measures the electrical properties of living tissue by passing a tiny, completely imperceptible electrical current between electrodes placed on the skin. Different tissue types resist electrical current differently: skin, fat, muscle, scar tissue, bone, and fluid all have distinct "impedance signatures" at different frequencies.
The sensor chip (AD5933, made by Analog Devices) generates a small AC signal that sweeps through frequencies from 1 kHz to 100 kHz and measures how the tissue responds at each frequency. The result is an "impedance spectrum" — a graph of resistance vs frequency that acts like a fingerprint for the tissue under the electrodes.
This is the same principle behind body composition scales that estimate your body fat percentage. Those scales send a tiny current through your feet and measure impedance — fat has high impedance (resists current), muscle has low impedance (conducts well), and water has very low impedance. Our sensor does the same thing, but locally, right at the point where the robot is about to work.
The electrodes are small gold-plated contact points (4 or 8 of them, depending on the design) arranged in a pattern around the working area. Gold is used because it's biocompatible (won't irritate skin), corrosion-resistant, and has excellent electrical conductivity. Spring-loaded pogo pins keep the electrodes in gentle contact with the skin surface.
Real-World Use:
Before tattooing each section, the bioimpedance sensor scans the skin to determine its type. Thin, hydrated skin (inner wrist) has very different impedance than thick, dry skin (elbow) or scar tissue (old injury). The AI uses this data to automatically adjust needle depth, speed, and force for each area. Without this sensor, the robot would use the same settings everywhere and produce inconsistent results — too deep on thin skin, too shallow on thick skin. With it, the robot adapts to each body like a human artist who can feel the difference between skin types with their hands.
Sensor 5: Strain Gauge — The Robot's Internal Stress Detector
Cost: $1-3 per gauge · Size: 5mm × 3mm foil · Weight: <0.1g
A strain gauge is the simplest and oldest sensor on this list, and in many ways the most fundamental. It's a thin metallic foil pattern bonded to a surface. When that surface bends, stretches, or compresses — even by a few microns — the foil deforms with it. Deformation changes the foil's electrical resistance. More stretch = more resistance. That's it. That's the entire principle. It was invented in 1938 and the basic concept hasn't changed because it didn't need to.
What makes strain gauges remarkable is their sensitivity. A typical gauge can detect deformations as small as 1 microstrain — which is 1 millionth of the original length. If you had a 1-meter steel bar, a strain gauge could detect it stretching by 0.001 millimeters. That's about 1/70th the width of a human hair. And it does this with a $1 piece of foil and a $3 amplifier.
Strain gauges are everywhere in engineering. They're in bathroom scales (measuring the deformation of a spring under your weight), in bridge monitoring systems (detecting structural stress), in aircraft wings (measuring flex during flight), and inside every force/torque sensor on the market (the ATI Mini45 is essentially a cleverly-arranged set of strain gauges on a machined metal structure).
In robotics, strain gauges are bonded directly to structural members of the robot's body or tools. They tell the robot when something is bending, twisting, or being loaded in a way it shouldn't be. They're the fundamental building block of force sensing — every other force sensor on this list is ultimately just a fancier arrangement of strain gauges or their conceptual descendants.
Real-World Use:
In our robot arm, strain gauges serve as the foundation of the DIY force/torque sensor option (Stanford CoinFT design). Six strain gauges arranged at specific angles on a 3D-printed flexure structure give us 6-axis force sensing for $40 instead of $3,000. We also use them on the spring-loaded pen mount — gauges on the mounting shafts tell us exactly how much the springs are compressed, which tells us how hard the pen is pressing against the skin. This is a backup measurement that cross-checks the main F/T sensor. If they disagree, something is wrong and the system pauses.
How All Five Work Together
No single sensor gives a robot "touch." Just like human touch isn't one sense — it's pressure, temperature, vibration, pain, and stretch all combined — robotic touch requires fusing data from multiple sensors into a unified understanding of what's happening at the point of contact.
Here's what it looks like when all five sensors are working simultaneously on our tattoo robot:
🔄 One Second of Robot Touch (1,000 sensor readings)
Force/Torque Sensor: "Needle is pressing into skin at 1.8N downward force, 0.1N lateral force. The lateral force suggests the needle is entering at a 3° angle — slightly off perpendicular. Correcting arm angle."
Pressure Film (fingertips): "Left finger gripping at 3.2N, right at 2.8N, top at 3.0N, bottom at 2.9N. Skin is stretched evenly. No slippage detected."
Accelerometer: "Needle vibration at 118 Hz, amplitude normal. No anomalous frequency spikes. Surface is normal skin tissue. Client is breathing — 0.3Hz oscillation detected, compensating arm position."
Bioimpedance: "Tissue under current position: normal dermis, hydration level moderate, no scar tissue detected. Impedance consistent with forearm skin. Maintaining current depth settings."
Strain Gauges: "Pen mount spring compression: 2.1mm. Consistent with 1.8N applied force from F/T sensor. Cross-check passed. No mechanical anomalies."
All of this happens 1,000 times per second. The AI fuses these data streams into a single real-time understanding: "The needle is at the correct depth, entering at the correct angle, the skin is properly stretched, the tissue is normal, and all mechanical systems are functioning correctly. Continue tattooing."
If any sensor reports something unexpected — a sudden force spike (client flinched), a vibration change (needle hit scar tissue), a pressure drop (finger lost grip), an impedance anomaly (tissue type changed), or a strain gauge disagreement (mechanical problem) — the robot pauses in under 1 millisecond. That's faster than a human blink. That's faster than any human tattoo artist could ever react.
What It All Costs
For $122, you can give a robot a complete sense of touch — force measurement in all six axes, distributed pressure on four fingertips, vibration analysis, tissue identification, and structural strain monitoring. The same capability in a commercial research robot costs $15,000-50,000. The sensors themselves are cheap. The hard part is the software that makes sense of the data — and that's what we build.
Frequently Asked Questions
Can a robot really "feel" as well as a human hand?
In some ways, better. A human can detect about 0.4 Newtons of force difference with their fingertips. An ATI F/T sensor detects 0.05N — eight times more sensitive. Accelerometers detect vibrations up to 3,200 Hz; humans top out around 800 Hz. Bioimpedance gives robots tissue information humans can't perceive at all. Where humans win is in the sheer density of sensors (17,000 in one hand) and the brain's ability to integrate it all effortlessly. Robots need explicit programming for every integration step.
Are these sensors safe to use on human skin?
Yes. The bioimpedance current is typically 200 microamps at most — you can't feel it. Medical devices like ECG monitors, body composition scales, and impedance tomography systems use the same principle and are FDA-approved for direct skin contact. The gold electrodes are biocompatible (used in medical devices worldwide). The other sensors don't directly contact skin — they're mounted on the robot's structure.
What happens if a sensor fails mid-operation?
Redundancy and cross-checking. The strain gauges on the pen mount cross-verify the F/T sensor. The accelerometer provides a second source of contact information. If any two sensors disagree beyond a threshold, the system pauses immediately. No single sensor failure can cause dangerous behavior because no single sensor is trusted alone.
Building Something That Needs to Feel?
We integrate touch sensing into robotic systems for manufacturing, security, medical devices, and custom applications. If your project needs a robot that interacts with the physical world, we can engineer the sensor system and write the AI software that makes sense of it.
Talk to Our Engineering Team →