Robot Contact Sensing: FSR Zones or a Capacitive Taxel Array?

Robot teams should choose the sensing architecture before they freeze the fingertip, gripper pad or contact surface. A few FSR zones can answer whether contact occurred and which area was loaded. A dense capacitive taxel array can support a spatial contact map, but it also brings more channels, mechanics, calibration and data processing. These are different jobs, not interchangeable sensor labels.

This distinction is timely. A July 31, 2026 original research paper on TacPrint describes a wearable fingertip with 24 capacitive taxels and a learned pipeline that estimates contact depth and location. The paper is useful because it shows what a spatial tactile system is designed to recover. It does not establish that every robot needs a dense array, and it does not describe an FSR product.

For a custom FSR pressure sensor, the practical question is narrower: does the equipment need discrete contact, relative load at known zones, or a detailed pressure image? Define that output first, then quote the sensing zones, actuator stack, readout circuit, tail and connector that can support it.

Start with the decision the robot must make

Write the control decision without naming a sensor. Examples include detecting first contact, confirming that both gripper fingers are loaded, distinguishing center contact from edge contact, checking whether a person pressed one of several padded zones, or reconstructing the shape and position of an object across a surface.

The first four decisions may be served by a small number of separate sensing zones if their positions are known and each channel has enough margin. The last decision usually needs a denser spatial sensor and an interpretation method. Adding more FSR zones does not automatically create a pressure image. Each zone still reports a local relative response shaped by its actuator, support, electronics and calibration.

Do not specify "high accuracy" as the output. State what error matters. A gripper may tolerate coarse contact location but cannot tolerate a missed edge contact. A protective panel may only need to confirm that load crossed a threshold. A teaching fingertip may need the contact centroid and shape. These acceptance conditions lead to different hardware.

Compare discrete FSR zones with a capacitive taxel array

An FSR zone is useful when a compact flexible part must detect contact or relative pressure at a defined location. Several zones can be printed into one outline and routed through a common tail or cable. The buyer still needs a readout circuit for each channel, a controlled actuator footprint and an installed calibration plan.

A capacitive taxel array is intended to sample many locations. The TacPrint research uses 24 channels beneath a silicone skin and applies a learned model to estimate a denser contact-depth map. That architecture supports a different information target from a small set of FSR zones. It also depends on the silicone geometry, array spacing, mounting, electronics, training data and model behavior.

Use a simple comparison during concept review:

• Choose discrete FSR zones when the decision is contact/no contact, relative load, threshold crossing, left-versus-right balance or a small known set of contact locations.
• Consider a dense tactile array when the decision needs contact position, boundary, shape or a spatial depth map across many possible locations.
• Do not expect an FSR to provide calibrated absolute force without system-level fixtures, electronics and calibration.
• Do not treat a dense array as a component-only shortcut. Channel count, data capture, model or algorithm behavior and final robot validation remain system responsibilities.

The most expensive mistake is selecting the dense option for a simple threshold task, or selecting a few zones and later asking software to recover spatial information the hardware never measured.

Define each FSR zone by actuator and load path

If discrete zones fit the task, draw the active areas against the actual contact surface. A zone should sit where force will enter the structure after tolerance shift, cover deflection and assembly variation. Its actuator may be a rubber pad, foam puck, plastic boss, gripper liner or compliant skin feature.

Record actuator width, shape, hardness and resting clearance. A narrow hard boss can concentrate force and produce a strong local response, but small lateral movement may miss the intended area. A wide compliant pad tolerates misalignment, yet it may spread load into a neighboring zone or leave preload after assembly. The rear support must remain flat enough that housing ribs, adhesive steps or fasteners do not load the sensor at rest.

For adjacent zones, specify the minimum separation and test cross-talk with the final actuator stack. Press one zone at the light, normal and high test levels while logging every channel. Then apply force between zones and at the edge. A layout that works under a centered bench indenter may fail when the installed contact patch is larger or angled.

This zone-and-actuator evidence cannot be copied from a generic sensor data sheet. It belongs to the robot's mechanical stack.

Plan channel count, readout and calibration together

Every added zone creates another signal path to read, filter and validate. Define whether the design uses separate analog inputs, a scanning arrangement or another circuit chosen by the equipment team. Record the supply condition, reference components, sampling rate and the unloaded baseline expected after assembly.

FSR response is affected by load history, dwell, actuator geometry and unit variation. Approve a repeatable decision margin rather than one resistance value from one loose sample. The test should include unloaded recovery, repeated contact, held load, off-center contact, temperature or environmental conditions that matter, and opening and refitting the housing.

For a dense capacitive system, the evidence package is different. Channel coupling, sampling synchronization, silicone or dielectric geometry, spatial reconstruction and model limitations become central. The recent TacPrint paper itself notes limits around boundary and shape reconstruction. That is a useful reminder that more channels and a learned map do not remove the need for task-specific validation.

The equipment maker owns the readout electronics, firmware thresholds, control response, safety logic and final robot performance. A sensor supplier can review the flexible sensor outline, zone layout, conductor routing, tail, cable and connector against the supplied mechanical and electrical inputs.

Keep the tail and connector out of the contact problem

The sensor outline is only one part of the RFQ. Mark the tail exit, first bend, cable transition, strain-relief point, service loop and connector envelope. A cable that pulls on the film can change how a zone sits on its support. A connector trapped behind a moving gripper joint can turn a good sensing layout into an assembly failure.

Route conductors away from screw bosses, hard stops and repeated flex regions. Keep the transition off the active contact surface. If the sensor enters a moving finger or curved pad, identify which bend is formed once during assembly and which region moves in service. A static tail bend and a repeated dynamic flex require different review.

At sample approval, photograph the sensor before the cover closes and again after the cable is restrained. Log all channels while the joint moves without contact. Any signal change from cable motion should be separated from a real contact event before thresholds are accepted.

Send an RFQ that states the information target

Start the RFQ with one sentence: "The robot must decide..." Then attach the mechanical and electrical evidence needed to support that decision.

Include:

• the required output, such as threshold contact, relative load, zone identity or spatial contact map;
• the number and location of proposed sensing zones;
• sensor outline, active-area dimensions and tolerance datums;
• actuator material, hardness, footprint, travel and resting clearance;
• support surface, adhesive, housing section, fastener and hard-stop locations;
• expected light, normal, overload and held-load conditions;
• channel readout concept, supply, reference circuit and sampling requirement;
• tail route, bend regions, cable length, strain relief and connector details;
• sample quantities and the mounted acceptance test for cross-talk, recovery and false triggers.

Use the Request Quote route to send the zone drawing and installed stack. The next decision is not "FSR or capacitive" in the abstract. It is whether the robot needs a few defensible contact decisions or a spatial tactile map, followed by hardware and validation evidence that matches that information target.