Force-Torque Sensor

A force-torque sensor is a transducer that measures the forces and torques (rotational forces) applied to an object, typically providing a multi-dimensional vector output of three orthogonal force components and three orthogonal torque components [8]. These sensors are a critical component in robotic systems, particularly for implementing force control, which allows a robot to interact with its environment by regulating the contact forces it exerts, rather than solely controlling its position [8]. The fundamental importance of these measurements stems from the concept of mechanical impedance, which describes how a structure resists motion when subjected to forces [1]. By directly measuring interaction forces, force-torque sensors enable robots to perform complex tasks requiring dexterity, adaptability, and safety, bridging the gap between purely position-controlled automation and intelligent physical interaction. The operational principle of most industrial force-torque sensors is based on measuring the strain (deformation) induced in a sensing element, often called a transducer body, by applied loads. This strain is converted into an electrical signal, typically via strain gauges arranged in Wheatstone bridge configurations, allowing for the precise calculation of the six force and moment components. Key characteristics of these sensors include their measurement range, resolution, accuracy, bandwidth, and stiffness. A major technical challenge in their use is compensating for parasitic effects, such as friction within the sensor's own mechanics or transmission elements, which can corrupt the force signal and degrade control performance [4]. Advanced [signal processing](/page/signal-processing "Signal processing is a fundamental engineering discipline...") techniques, including adaptive filtering, are employed to isolate the desired force measurements from noise and structural vibrations [5]. Common types include sensors based on piezoelectric, capacitive, or optical principles, with designs ranging from compact, wrist-mounted units for robotic end-effectors to larger multi-axis load cells for industrial machinery. Force-torque sensors have wide-ranging applications that underscore their significance in modern technology. In robotics, they are essential for advanced manufacturing tasks like precise assembly, grinding, deburring, and force-guided insertion [8]. They enable impedance control strategies, where the robot's dynamic response to contact is modulated to achieve stable and compliant interaction, a foundational approach for manipulation [2]. Beyond manufacturing, these sensors are pivotal in research on physical human-robot interaction, enabling collision detection and safe collaboration [3]. In machining processes like milling, monitoring cutting forces is crucial for chatter recognition, which affects surface quality and tool life [6]. Furthermore, force-torque sensing is a cornerstone for robotic learning, where variable impedance control models allow robots to adapt their physical behavior based on interaction forces to follow desired trajectories or learn tasks from demonstration [7]. Their role extends to aerospace, biomedical devices, and rehabilitation robotics, making them indispensable for any system requiring intelligent physical interaction with the world.

Overview

A force-torque sensor (F/T sensor) is a transducer that measures the multi-axis forces and torques (moments) applied to its sensing element, providing a complete description of the mechanical interaction at a point in space. These sensors are fundamental to enabling robots and automated systems to interact intelligently and safely with their environment by providing direct feedback on contact forces, moving beyond purely position-based control. The core function is to convert the six components of a wrench vector—three orthogonal forces (Fx, Fy, Fz) and three orthogonal torques (Tx, Ty, Tz)—into a corresponding electrical signal, typically through the measurement of strain in a precisely engineered elastic structure [14].

Fundamental Operating Principle and Mechanical Impedance

The operation of a force-torque sensor is intrinsically linked to the concept of mechanical impedance, which is a measure of how a structure resists motion when subjected to an external force [13]. In essence, the sensor's elastic body (often called a flexure) is designed to have a known, controlled impedance. When external forces and torques are applied, they cause this structure to deform minutely. This deformation, or strain, is measured by an array of strain gauges—typically arranged in Wheatstone bridge configurations for sensitivity and temperature compensation—bonded to strategic locations on the flexure. The measured strain is directly proportional to the applied load, following Hooke's Law within the sensor's elastic range. The fundamental idea builds from the fact that mechanical impedance is a measure of how a structure resists motion when subjected to forces; the sensor is designed to have a precise, measurable impedance that translates force into a quantifiable displacement or strain [13]. By analyzing the differential signals from multiple strain gauge bridges, the sensor's electronics can resolve the individual force and torque components.

Key Specifications and Performance Metrics

The performance of a force-torque sensor is characterized by several critical specifications:

• Number of Axes: While six-axis (Fx, Fy, Fz, Tx, Ty, Tz) is standard for complete wrench measurement, some applications use one-, two-, or three-axis sensors for specific force or torque measurements [14].
• Measurement Range (Full-Scale Output): The maximum force and torque values the sensor can measure accurately, specified per axis (e.g., Fx,y: ±500 N, Fz: ±1000 N, Tx,y,z: ±50 Nm). Operating beyond this range risks permanent damage or non-linear output.
• Resolution: The smallest detectable change in force or torque, limited by electrical noise in the signal conditioning system. High-resolution sensors can detect changes as small as 0.01% of the full-scale range.
• Accuracy: The closeness of the measured value to the true applied load, often expressed as a percentage of full scale (e.g., ±0.5% F.S.). It encompasses non-linearity, hysteresis, and non-repeatability errors.
• Stiffness: The spring constant of the sensor's elastic element, determining how much it deflects under load. High stiffness is often desired to minimize deflection and maintain system positional accuracy, but it involves a trade-off with sensitivity.
• Natural Frequency: The resonant frequency of the sensor structure, which dictates the bandwidth for dynamic force measurements. A high natural frequency (often > 1 kHz) is necessary for capturing rapid force transients without aliasing or resonance amplification [14].
• Crosstalk (Inter-Axis Coupling): The error in one axis caused by a load applied on another axis, expressed as a percentage (e.g., < 0.5%).
• Overload Rating: The maximum load the sensor can withstand without mechanical failure, typically 150-300% of the rated measurement range.

Sensor Designs and Technologies

Several transducer technologies are employed, with strain gauge-based sensors being the most prevalent due to their robustness, linearity, and high stiffness.

• Strain Gauge-Based Sensors: These use metallic foil or semiconductor strain gauges. A common design is the Maltese cross or similar monolithic transducer, where strain gauges are mounted on multiple sensing beams within a single piece of metal (like aluminum or stainless steel) to decouple force and torque components. Another widespread design is the Stewart Platform-style sensor, which uses a hexapod structure with six individual single-axis load cells; the spatial arrangement allows computation of all six wrench components from the six load cell readings [14].
• Optical and Capacitive Sensors: Emerging designs use optical encoders or capacitive plates to measure the microscopic deflections of an internal flexure, offering high resolution and immunity to electromagnetic interference.
• Piezoelectric Sensors: These use quartz or ceramic crystals that generate an electric charge proportional to applied stress. They are excellent for measuring very high-frequency dynamic forces but are not suitable for static loads due to charge leakage.

Signal Processing and Calibration

The raw millivolt-level signals from strain gauge bridges require sophisticated conditioning. This includes amplification, analog-to-digital conversion, and digital filtering. A critical process is sensor calibration, which establishes the mathematical relationship between the six voltage outputs and the applied wrench. This is represented by a 6x6 calibration matrix, C. The relationship is given by: w = C * v, where w is the 6x1 wrench vector (forces and torques) and v is the 6x1 vector of conditioned voltage readings from the transducer bridges [14]. Determining the matrix C requires applying a large set of known, pure loads (using calibration machines) and performing a least-squares regression. Advanced calibration also compensates for temperature effects and non-linearities. The processed digital force-torque data is then transmitted via standard industrial communication protocols like EtherCAT, Ethernet/IP, or analog voltages to the robot controller.

Role in Impedance and Force Control

Force-torque sensors are the enabling hardware for implementing advanced control paradigms like impedance control and direct force control, which are essential for physical interaction. In impedance control, the sensor measures the interaction forces between the robot's end-effector and the environment. The controller then modulates the robot's motion to achieve a desired dynamic relationship—the target impedance—between the measured force and the resulting position or velocity error [13]. A standard impedance interaction model is defined as M(ẍ - ẍ_d) + B(ẋ - ẋ_d) + K(x - x_d) = F_ext, where M, B, and K are the desired inertia, damping, and stiffness matrices respectively, x_d and x are the desired and actual positions, and F_ext is the measured interaction force [13]. This allows the robot to behave like a programmable spring-damper-mass system, compliantly reacting to contacts. In direct force control, the sensor provides feedback for a closed-loop controller that directly regulates the interaction force to a desired setpoint, crucial for tasks like polishing, assembly, and precise material testing [14].

History

The development of force-torque sensors is intrinsically linked to the broader evolution of robotics and automation, where the need to measure interaction forces between a machine and its environment became paramount. The conceptual foundation for these sensors can be traced to the study of mechanical impedance, which is a measure of how a structure resists motion when subjected to an applied force. This fundamental principle underpins the relationship between force, displacement, and the dynamic behavior of mechanical systems, forming the theoretical basis for sensing both static and dynamic loads.

Early Foundations and Conceptualization (Pre-1970s)

The earliest precursors to modern force-torque sensors were simple load cells and strain gauges used in mechanical testing and industrial weighing applications. These devices, which measured uniaxial force through the deformation of a mechanical element, provided the initial technological building blocks. However, the specific need for a compact, multi-axis sensor capable of measuring both forces and moments (torques) about multiple axes emerged directly from the field of robotics and advanced manufacturing. Researchers recognized that for a robot to interact intelligently with its environment—beyond simple pre-programmed motions—it required sensory feedback about contact forces. This need drove the initial research into designing structures that could transduce multi-dimensional wrench vectors (combinations of forces and torques) into measurable electrical signals, often by strategically placing strain gauges on a specially designed elastic element or "flexure."

Emergence and Initial Applications (1970s-1980s)

The 1970s and 1980s marked the period of active research and development that led to the first true multi-axis force-torque sensors. Pioneering work in robotic manipulation and control theories created the demand for this technology. A key driver was the development of impedance control, a robotic control strategy where the manipulator is programmed to exhibit a specific dynamic relationship between its position and the contact force. Implementing such advanced control schemes was impossible without a sensor providing direct, real-time measurement of the interaction wrench. Early applications were primarily in academic and industrial research labs, where these sensors were used for tasks such as:

• Robot assembly, where parts must be mated with controlled forces to prevent jamming or damage. - Mechanical characterization of materials and structures. - Fundamental research in robotic control algorithms beyond pure position control. The modeling and identification of these sensor systems, including their stiffness matrices and dynamic responses, became an active area of research, with numerous methods proposed in the literature to improve accuracy and compensate for cross-coupling between axes.

Refinement and Widespread Industrial Adoption (1990s-2000s)

By the 1990s, force-torque sensor technology had matured sufficiently for broader commercial and industrial use. Advancements in precision machining, strain gauge technology, and signal conditioning electronics led to sensors with improved accuracy, reliability, and robustness. Their integration became critical in sophisticated automated manufacturing processes. A prime example is in robotic machining, such as milling and deburring, where the sensor allows the robot to compensate for tool wear, part misalignment, and path inaccuracies by adapting its position based on measured contact forces. To ensure stable machining operations, research into chatter identification became relevant. For instance, methods based on dynamic wavelet packet decomposition (WPD) were proposed from the perspective of signal processing to avoid chatter by detecting its onset through force-torque signals [18,19,20]. This era also saw the standardization of interfaces and the emergence of several commercial manufacturers specializing in multi-axis sensors, cementing their role as a key enabling technology for force-controlled robotics.

Integration into Advanced Robotic Systems (2000s-Present)

The 21st century has been defined by the integration of force-torque sensors into increasingly complex and sensitive robotic systems, most notably in medical robotics. This application has placed extreme demands on sensor performance, requiring not only high precision but also compactness, sterility, and reliability. In robot-assisted surgery, force-torque sensing addresses a critical limitation of earlier systems: the lack of haptic feedback for the surgeon. As noted in a meta-analysis, the proposed solutions for providing haptic feedback vary greatly in terms of the methods, how the feedback is provided to the surgeon, and their intended fields of application [15]. The integration of force sensing is a fundamental component of many of these solutions. The clinical implementation and benefits of this technology have been demonstrated in various surgical fields. For example, in thoracic surgery, the introduction of force feedback has been systematically analyzed. Studies have shown that force feedback versions of instruments commonly used in thoracic surgery, including the Cadiere Forceps, Fenestrated Bipolar Forceps, Maryland Bipolar dissector, and Needle Driver, became available and were integrated into surgical practice [16]. This technological evolution allows surgeons to perceive tissue interaction forces directly, which is hypothesized to improve surgical precision, reduce tissue trauma, and shorten learning curves for complex procedures. The benefits of such haptic feedback, however, are moderated by factors including the specific surgical task and the implementation method, as analyzed in the literature [15]. Concurrently, in industrial settings, force-torque sensors have become central to the development of collaborative robots (cobots). These robots are designed to work alongside humans, and force sensing enables essential safety and functionality features such as:

• Power and Force Limiting (PFL): Direct measurement of contact forces to ensure they remain below injury thresholds.
• Manual Guidance: Allowing an operator to physically teach the robot a path by moving its arm, with the sensor measuring the applied guiding forces.
• Advanced Assembly: Performing delicate insertions and complex assemblies with adaptive control. Building on the concept of overload rating mentioned previously, modern sensors are designed to withstand significant accidental overloads while maintaining their calibration, a necessity for both surgical and collaborative environments where unexpected contact is possible. Furthermore, as noted earlier, a high natural frequency is critical in these dynamic applications to capture rapid force transients accurately without signal distortion. The historical trajectory of force-torque sensors illustrates a path from a specialized research instrument to a fundamental component in advanced robotics. Their development has been consistently driven by the need to bridge the gap between the digital control world of machines and the physical world of mechanical interaction, enabling robots to perform tasks with a sensitivity and adaptability that approaches human capability.

Principles

The fundamental operation of force-torque sensors is governed by the principles of elasticity, strain transduction, and multi-axis load decomposition. These sensors convert applied mechanical loads into measurable electrical signals through the deformation of a precisely engineered elastic structure, typically made from high-strength aluminum alloys (e.g., 7075-T6) or managing steel [17]. The core principle involves measuring the strain induced in this structure, which is directly proportional to the applied force or torque according to Hooke's law within the material's elastic limit. For a simple uniaxial case, the relationship is σ = Eε, where σ is the stress in pascals (Pa), E is the Young's modulus of the material (e.g., ~71 GPa for aluminum 7075), and ε is the strain (dimensionless, typically measured in microstrain, με) [20]. The complete six-degree-of-freedom (6-DoF) load vector F = [F_x, F_y, F_z, T_x, T_y, T_z]^T is resolved from the strain field measured at multiple points on the structure using a network of strain gauges or other transducers.

Elastic Structure and Load Decomposition

The elastic element, or transducer body, is designed to provide a known, deterministic relationship between applied loads and resulting strain patterns at specific locations. Common designs include maltese cross, double-walled cylinder, and shear beam configurations. The governing equation for the sensor's output is a linear transformation: V = C

• F, where V is a vector of output voltages from the sensing elements (typically in millivolts per volt of excitation, mV/V), and C is the sensor's calibration matrix (units of, e.g., mV/V/N or mV/V/Nm) [17]. This matrix is determined through a rigorous calibration process applying known pure loads along each axis. The strain ε at a gauge location for a combined load is a superposition: ε_total = Σ (S_i * F_i), where S_i is the sensitivity coefficient for the i-th load component. Cross-axis sensitivity, or crosstalk, is a critical parameter kept below 1-5% in precision sensors through symmetric mechanical design and algorithmic compensation using the inverse of the calibration matrix: F = C⁻¹
• V [20].

Signal Generation and Conditioning

Strain is most commonly measured using foil strain gauges arranged in Wheatstone bridge configurations. A full bridge for one channel consists of four active gauges, with two gauges experiencing tension and two compression under load, maximizing output and providing temperature compensation. The output voltage ΔV of a Wheatstone bridge with excitation voltage V_ex is approximately ΔV = (V_ex * GF * ε) / 4, where GF is the gauge factor (typically ~2.0 for metallic foils). For a sensor with a rated capacity of, for example, 100 N in force and 5 Nm in torque, a full-scale bridge output might range from 1.5 to 3 mV/V [17]. This low-level signal requires amplification by instrumentation amplifiers with high common-mode rejection ratio (CMRR > 100 dB) and low noise (< 1 μV RMS). Subsequent analog-to-digital conversion occurs at 16- to 24-bit resolution with sampling rates from 1 kHz to over 10 kHz, necessary for capturing dynamic processes while avoiding aliasing, as noted earlier regarding the need for a high natural frequency [1].

Modeling, Identification, and Advanced Processing

Accurate interpretation of the raw voltage signals requires sophisticated modeling and identification methods to account for non-ideal behaviors. These methods, extensively proposed in the literature, address issues such as hysteresis, creep, and temperature drift [4]. A dynamic model often includes terms for the sensor's mass and damping, extending the static calibration matrix to a transfer function G(s) relating force input to voltage output in the Laplace domain. Parameter identification techniques, such as least-squares estimation or recursive algorithms, are used to fit these models to experimental data [4]. Furthermore, advanced signal processing techniques are employed for specific applications. For instance, in machining, to avoid chatter, a milling chatter identification method based on dynamic wavelet packet decomposition (WPD) has been proposed from the perspective of signal processing [6]. This method analyzes the high-frequency force-torque signals to detect the characteristic frequency patterns associated with the onset of chatter instability.

Integration into Control Architectures

The primary utility of force-torque sensors is realized when their measurements are integrated into closed-loop control systems, enabling robots to interact intelligently with their environment. This demands the use of advanced interaction methodologies based on impedance control, where the robot's dynamic behavior at the end-effector is modulated to mimic a mass-spring-damper system [13]. The measured force F_ext is used in the impedance control law: M_d(ẍ

• ẍ_d) + B_d(ẋ
• ẋ_d) + K_d(x
• x_d) = F_ext, where M_d, B_d, K_d are the desired inertia, damping, and stiffness matrices, and x, ẋ, ẍ are the actual position, velocity, and acceleration [13]. In addition to tracking motion planning trajectories, control techniques that reason about contact forces are essential for tasks requiring physical interaction [18]. These techniques can benefit from the use of different methods of force control strategies, such as hybrid position/force control or parallel force/position control [1]. For example, Mario Farrugia designed a generic controller for flexible assembly using a network of six transputers to implement such generic control strategies [19]. The sensor provides the critical feedback for these strategies, allowing the controller to maintain a desired force profile (e.g., 10 ± 0.5 N for a deburring operation) while accommodating positional uncertainties in the workpiece.

Types

Force-torque sensors can be systematically classified along several dimensions, including their fundamental operating principle, mechanical structure, and the specific control methodologies they enable. This classification is essential for selecting appropriate sensors for applications ranging from precision robotic assembly to human-robot collaboration.

By Operating Principle

The underlying physical principle used to transduce force and torque into a measurable electrical signal defines a primary categorization.

• Strain Gauge-Based Sensors: This is the most prevalent type, where foil or semiconductor strain gauges are bonded to a mechanically deforming structure, often called a transducer body. The applied load induces strain, altering the gauge's electrical resistance in a measurable way via a Wheatstone bridge circuit. The relationship between strain (ε) and resistance change (ΔR/R) is given by the gauge factor (GF): ΔR/R = GF · ε. These sensors are prized for their high linearity, robustness, and ability to achieve the high natural frequency necessary for dynamic applications [18]. They are typically integrated directly into the robot's mechanical structure near the wrist or flange.
• Piezoelectric Sensors: These utilize materials like quartz or specialized ceramics that generate an electric charge proportional to an applied mechanical stress. They excel at measuring highly dynamic forces and torques with an exceptionally wide bandwidth, often exceeding 5 kHz, making them ideal for impact detection or vibration analysis. However, they are inherently unsuitable for measuring static loads due to charge leakage. Their signal conditioning requires specialized charge amplifiers [24].
• Optoelectric Sensors: This category uses optical principles, such as measuring the deflection of a structure via changes in light intensity passing through an aperture or shifts in the position of a light spot on a photodiode array. Some advanced designs use Fiber Bragg Gratings (FBGs), where strain alters the reflected wavelength of light in an optical fiber. These sensors offer high immunity to electromagnetic interference and can be advantageous in harsh environments [25].
• Capacitive Sensors: These measure force-induced displacements by detecting changes in capacitance between conductive plates. They can offer high resolution and stability but often have a more limited measurement range compared to strain gauge types. Their compact form factor can be beneficial in space-constrained applications.

By Mechanical Structure and Design

The physical design of the transducer body, which deforms under load, determines key performance characteristics like sensitivity, cross-talk, and overload protection.

• Cross-Beam Sensors: Featuring a simple cruciform or "cross" structure, these are common for compact, low-cost six-axis sensors. Strain gauges are mounted on the beams connecting a central hub to an outer ring. Their design offers a good balance of stiffness and sensitivity but can exhibit higher cross-talk between axes compared to more complex designs.
• Maltese Cross Sensors: A more advanced evolution of the cross-beam design, incorporating additional mechanical decoupling features to minimize cross-talk. The structure often includes multiple bending and shear elements strategically placed to isolate the response to individual force and torque components.
• Column-Type Sensors: These use a central column or multiple columns as the primary elastic element. They are often characterized by very high stiffness and overload capacity, making them suitable for heavy-duty industrial applications like machining force monitoring. The modeling and identification of such structures for control purposes is an active area of research [18,19,20].
• Ring-Type Sensors: Utilizing a circular or octagonal ring structure, these designs can provide excellent symmetry and uniform sensitivity. They are frequently used in applications requiring precise multi-axis measurement with minimal hysteresis.
• Platform Sensors (Multi-Axis Load Cells): While often considered a distinct category, platform-style sensors that measure forces in three orthogonal axes (Fx, Fy, Fz) can be viewed as a subset of force-torque sensors. They typically lack the direct measurement of torque about these axes but form the basis for more complex six-axis designs.

By Integration and Control Methodology

Sensors are also classified by their role within the robotic control architecture, which dictates their required specifications and signal processing.

• Direct Force Control Sensors: These sensors provide the primary feedback signal for force-controlled loops. They require high bandwidth, low noise, and excellent linearity. The raw or minimally processed signal is used in control laws like hybrid position/force control or impedance control. To limit the negative effect of measurement noise, the signal is often conditioned using a low-pass filter or a moving average filter before being fed to the controller [19].
• Impedance Control Sensors: In impedance control, the sensor measures the interaction forces between the robot and its environment to modulate the robot's apparent mechanical impedance (the dynamic relationship between force and motion). This approach does not explicitly control force but rather dictates how the robot reacts to contact. The fundamental idea builds from the concept that mechanical impedance is a measure of how a structure resists motion when subjected to a force. Advanced implementations use adaptive impedance control, where control parameters are adjusted in real-time based on sensor data to improve performance [22].
• Collision Detection and Safety Sensors: For collaborative robotics (cobots), sensors must reliably detect unexpected contact to ensure human safety. This application prioritizes very high reliability and fast response time over extreme precision. Algorithms process force-torque data to distinguish intentional contact from a dangerous collision. Research in this domain evaluates biomechanical limits, such as force pain thresholds, to define safe robot behavior parameters [24].
• Haptic Feedback Sensors: In teleoperation systems, a force-torque sensor on the remote robot's end-effector measures interaction forces. This data is then transmitted to a master device that delivers haptic feedback to the human operator, stimulating somatic receptors to generate a sensation of touch [25]. This requires sensors with high fidelity and low latency to create a realistic tactile experience.

Standards and Performance Classification

While no single universal standard defines all sensor types, several international standards guide their testing and specification. Key performance parameters, beyond those mentioned previously, include:

• Nonlinearity: The maximum deviation of the calibration curve from a best-fit straight line, typically expressed as a percentage of Full Scale Output (FSO).
• Hysteresis: The maximum difference in output at any load value when the load is approached from increasing and decreasing directions.
• Cross-Talk (Interaxis Coupling): The signal output on one measurement axis when a load is applied exclusively to another axis. High-performance sensors aim for cross-talk values below 1-2% of FSO.
• Temperature Sensitivity: The change in zero-point offset and sensitivity due to temperature variations, specified in %FSO/°C or με/°C. The choice of sensor type is a critical design decision that balances these performance characteristics against application requirements for precision, dynamics, robustness, and cost. As noted earlier, the sensor's integration into the low-level control firmware running directly on the robot arm is paramount for achieving high-performance force-responsive behaviors [18].

Characteristics

The operational characteristics of force-torque sensors define their performance in real-world applications, encompassing signal processing, control system integration, safety compliance, and domain-specific implementation requirements. These characteristics bridge the fundamental sensor properties, such as the high natural frequency and overload rating noted earlier, with their practical utility in robotic and automated systems.

Signal Processing and Filtering

Raw measurements from the sensing elements require significant electronic conditioning to produce stable, usable signals. To limit the negative effect caused by potential force measurement errors, a low-pass filter, and a moving filter are commonly used to smooth the measured force [7]. This processing is critical because electrical noise, mechanical vibrations, and quantization errors from analog-to-digital converters can introduce high-frequency artifacts into the signal. The design of these filters involves a fundamental trade-off: aggressive filtering improves signal stability but introduces phase lag and reduces the effective bandwidth, potentially masking rapid force transients that the sensor's physical structure is capable of detecting. The cutoff frequency for these filters is typically selected to be well below the sensor's natural frequency to avoid exciting structural resonances while preserving the bandwidth necessary for the control task. For instance, in dynamic applications like collision detection, a higher cutoff frequency is maintained, whereas in steady-state force control tasks like precision assembly, more aggressive filtering can be applied to achieve sub-Newton resolution.

Control System Integration and Implementation Level

Force-torque sensors are integrated into control architectures at various levels, each with distinct implications for performance and complexity. Typically the low-level controllers are implemented in the firmware that runs directly on the robot arm (or its control cabinet) [25]. This implementation close to the hardware minimizes communication latency, which is paramount for stability in high-gain force control loops or for triggering safety stops upon collision detection. In such architectures, the sensor's analog signals are digitized by a dedicated electronics unit, which may perform initial filtering and scaling before transmitting data via a high-speed fieldbus (e.g., EtherCAT, PROFINET) to the robot's main controller. The controller then uses this data in its servo loop, which may operate at frequencies between 500 Hz and 8 kHz. An alternative architecture uses an external sensor interface box that communicates with a separate PC-based controller; this offers more flexibility for algorithm development but can introduce additional communication delays that must be accounted for in the control law design.

Safety and Compliance Standards

The deployment of force-torque sensors, particularly in collaborative robotics (cobots), is governed by stringent international safety standards. As a preliminary measure to address safety concerns, the standard ISO/TS 15066 was established to provide safety guidelines for the collaborative operation of industrial robots [24][11]. This technical specification supplements the broader machinery safety standard ISO 10218 and provides crucial data on pain thresholds for various human body regions, which directly informs the design of force-limited control systems using these sensors. A cobot's control system uses real-time force-torque feedback to ensure that any contact force exerted on a human coworker remains below these biologically defined thresholds. Furthermore, in specific applications such as medical devices, compliance extends to biological evaluation standards. For example, the U.S. Food and Drug Administration references ISO 10993-1 for the biological evaluation of medical devices, which may encompass robotic surgical systems or rehabilitation devices that incorporate force sensing to interact with patients [14]. Adherence to these standards is not merely regulatory but fundamentally shapes sensor characteristics, necessitating features like functional safety (FuSa) certification (e.g., SIL 2/PL d), redundant signal paths, and built-in self-test diagnostics to ensure fail-safe operation.

Application-Specific Performance Requirements

The demanded characteristics of a force-torque sensor vary dramatically across different fields, dictating design priorities.

• Robotic Assistance and Rehabilitation: In devices like robotic walkers or exoskeletons, the sensor must reliably measure the interactive forces between the human and the machine to provide appropriate assistance. Research on a robotic walker for slope mobility demonstrated that suitable assistive forces can be provided by a system with a simple mechanical structure and control method, a design philosophy that often favors robust, cost-effective sensors with moderate bandwidth over ultra-high precision [7]. The key characteristic here is long-term reliability and environmental robustness, as the sensor must perform consistently in unpredictable, daily-life environments.
• Contact-Rich Manipulation and Exploration: Advanced robotic manipulation in unstructured environments, such as assembly, deburring, or plug insertion, requires sensors that enable sophisticated force-guided strategies. As explored in research on FORCE-Guided Exploration for Robust Contact-Rich Manipulation (FORGE), the sensor must provide high-fidelity, low-noise data to support exploration algorithms that reason about contact geometry and uncertainty [9]. For tasks like robotic deburring, the sensor enables the end-effector to keep constant speed and controlled force along a contour, even as part geometry or burr thickness changes [10]. This demands excellent resolution at lower force ranges, low hysteresis, and minimal temperature drift to maintain process consistency.
• Locomotion and Dynamic Balance: In legged robotics, force-torque sensors in the feet or legs are critical for state estimation and balance recovery. They measure ground reaction forces, which are used to calculate the center of pressure and zero-moment point—essential parameters for dynamic gait control. Research into balance recovery for a quadruped robot relies on accurate, high-bandwidth force data to make rapid adjustments to leg placement and body posture [8]. The sensors in these applications must withstand high-impact shocks and have a very high overload rating, as noted previously, to survive unexpected footfalls or jumps.
• Haptic Interaction and Teleoperation: In haptics, a force-torque sensor often acts as the input device, measuring the forces applied by a human operator to control a remote or virtual robot. The work of researchers like Yeongmi Kim in haptic displays emphasizes the need for sensors with extremely low friction and inertia to provide a transparent feel, alongside a wide dynamic range to capture both subtle exploratory forces and strong grasps [25]. Low axial stiffness can be a desirable trait in this context to allow for natural human motion, contrasting with the high stiffness typically sought in industrial robot wrist sensors. The convergence of these characteristics—real-time signal processing, deeply embedded control integration, adherence to safety protocols, and tailoring to application physics—defines the functional envelope of a modern force-torque sensor, transforming it from a measurement instrument into a core component of an intelligent, interactive mechanical system.

Applications

Force-torque sensors are critical components that enable robots and automated systems to interact intelligently and safely with their environment by providing direct measurement of contact forces. Their applications span from high-precision industrial assembly to delicate surgical procedures and advanced physical human-robot interaction. The data from these sensors is fundamental to implementing force control strategies, which, as noted earlier, operate within servo loops at frequencies between 500 Hz and 8 kHz [10].

Industrial Automation and Robotics

In industrial settings, force-torque sensors are indispensable for tasks requiring contact and precise force application. A primary application is in robotic assembly, where sensors enable compliant motion for inserting components, such as a peg into a hole, by allowing the robot to respond to contact forces and moments, preventing jamming and part damage [12]. They are also central to force-controlled grinding, polishing, and deburring, where maintaining a consistent contact force against a workpiece with complex geometry is essential for quality and repeatability. Beyond manufacturing, these sensors are vital for automated testing and quality inspection, measuring parameters like switch actuation force, button press feedback, and seal integrity. Safety is another critical domain; integrating force-torque sensing allows for the implementation of collision detection and reaction strategies, contributing to safer human-robot collaboration, a concern addressed by standards like ISO 10218-1 for industrial robot safety [30]. Major industrial automation providers, such as ABB, incorporate this technology into their robotic solutions for these advanced applications [12].

Medical and Surgical Robotics

The field of robot-assisted surgery (RAS) represents a major application area where force feedback is crucial yet challenging to implement. A significant drawback of current RAS systems is the loss of haptic sensation, which deprives surgeons of a natural and critical source of information about tissue properties and interaction forces during procedures [15]. This limitation is noted in various surgical specialties; for instance, in robot-assisted thoracic surgery (RATS), the lack of tactile feedback remains an important constraint on surgical capability [16]. Similarly, during procedures like robot-assisted needle insertion, the operator cannot directly feel the operation because the feedback from both surgeon-instrument and instrument-tissue interactions is lost, complicating tasks like vessel puncture detection or tissue layer differentiation [26]. Force-torque sensors, often mounted at the surgical instrument's wrist or within the manipulator, aim to address this by providing indirect force feedback to the surgeon through visual or vibrotactile cues, or by enabling autonomous force-limiting safety functions. The biological evaluation of such medical devices, including sensor components that may contact tissue, is governed by international standards like ISO 10993-1 [28].

Rehabilitation and Assistive Robotics

In rehabilitation robotics, force-torque sensors are key to implementing adaptive and patient-specific therapy. Traditional rehabilitation robots often cannot automatically modify training parameters, such as assistive force or resistance, to match the evolving rehabilitation status of a patient's limb [27]. By measuring the interaction forces between the robot and the patient, force-torque sensors enable impedance control and admittance control strategies. These allow the robot to dynamically adjust its behavior—from providing full guidance for a weakened limb to offering compliant resistance for strength training. This real-time adaptation, based on direct force measurement, is essential for applying effective and personalized rehabilitation protocols that can improve therapeutic outcomes [27]. Furthermore, such sensors are integral to advanced prosthetic limbs, enabling intuitive control via myoelectric signals and providing grip force feedback to the user.

Research and Advanced Robotic Systems

Force-torque sensors are foundational tools in robotics research, particularly in areas focused on dynamic interaction. They are essential for sim-to-real transfer in machine learning, where policies trained in simulation are deployed on physical robots. Accurate force sensing provides the real-world grounding necessary to close the reality gap in challenging domains like dexterous manipulation and legged locomotion [12]. In legged robotics, sensors mounted at the feet or ankles measure ground reaction forces (GRF), which are critical for estimating center of pressure, maintaining balance, and implementing dynamic gait controllers. For dexterous manipulation with robotic hands, fingertip or wrist-mounted sensors provide the data needed to modulate grip force, manipulate fragile objects, and perform complex in-hand manipulation tasks. Research into physical human-robot interaction (pHRI) also relies heavily on these sensors to ensure safe and natural contact, from collaborative industrial tasks to more nuanced social interactions.

Physical Human-Robot Interaction (pHRI)

Beyond industrial collaboration, force-torque sensors enable sophisticated social and assistive physical interaction between robots and humans. A canonical example is the robotic handshake. Research into this gesture seeks to replicate the nuanced force profiles, timing, and compliance observed in human-to-human handshakes, which is a culturally significant greeting ritual [29]. Achieving a natural handshake requires high-bandwidth sensing and control to manage the interactive forces and impedance smoothly. More broadly, in assistive and social robotics, these sensors allow a robot to detect external forces applied by a human—such as a guiding push or pull—and respond compliantly. This capability is essential for robots that physically assist the elderly or individuals with mobility impairments, including devices that help navigate challenging terrains like uphill and downhill roads, which can pose significant difficulty and danger [Key Points]. The sensor data enables the robot to understand the user's intent and provide appropriate support while maintaining safety.

Aerospace, Automotive, and Testing

In aerospace applications, force-torque sensors are used in wind tunnel testing to measure aerodynamic loads on scale models and in structural testing to apply and monitor precise loads on airframe components. The automotive industry employs them extensively in research and development, particularly for testing components like steering systems (measuring steering wheel torque), brake pedals, and switches. They are also crucial in vehicle durability testing, where they measure loads experienced by suspension components and chassis on simulated road profiles. Furthermore, they play a role in biomechanics research for analyzing gait and sports performance by measuring forces between the foot and the ground or between an athlete and equipment.

Considerations

The selection, integration, and application of force-torque sensors involve critical trade-offs and technical challenges that extend beyond basic performance specifications. These considerations are paramount for engineers and system designers to ensure reliable operation, accurate measurement, and long-term stability in demanding environments.

Integration and System-Level Challenges

Integrating a force-torque sensor into a robotic or automated system introduces complexities that can significantly impact overall performance. A primary concern is the alteration of the system's structural dynamics. The sensor, acting as a compliant element between the robot's wrist and its end-effector, lowers the overall stiffness of the manipulator [1]. This reduction in stiffness can decrease the natural frequency of the robot's end-point, potentially leading to vibrations and instability, especially during high-speed movements or when interacting with stiff environments [2]. Compensating for this effect often requires careful tuning of the control system's gains and may limit the maximum achievable bandwidth of force-controlled operations [3]. Electrical integration presents another layer of challenge. The low-level analog signals from strain gauges are highly susceptible to electromagnetic interference (EMI) and noise, which can originate from motor drives, power cables, or wireless equipment [4]. Proper shielding of sensor cables, the use of twisted-pair wiring, and physical separation from noise sources are essential practices. Furthermore, the signal conditioning electronics, including amplifiers and analog-to-digital converters (ADCs), must be carefully selected to match the sensor's output characteristics. The resolution of the ADC is particularly critical; for a sensor with a 500 N full-scale output and a desired resolution of 0.1 N, the ADC must resolve at least 1 part in 5000, necessitating a minimum of 13-bit resolution, though 16-bit or 24-bit ADCs are common in high-precision applications to provide ample margin [5].

Environmental and Operational Factors

Force-torque sensors are sensitive to a range of environmental conditions that can induce measurement errors. Temperature variation is the most significant factor, causing two primary effects: a change in the gauge factor of the strain gauges and thermal expansion/contraction of the sensor's metal body, which induces apparent strain [6]. High-quality sensors incorporate temperature compensation networks, often using dummy gauges in a Wheatstone bridge configuration, to nullify these effects. However, rapid thermal transients or large thermal gradients across the sensor body can still lead to temporary drift until thermal equilibrium is re-established [7]. Mechanical mounting and load introduction are fundamental to measurement accuracy. Off-center or non-axial loading can induce cross-talk errors, where a force applied along one axis produces an erroneous signal on another axis [8]. Mounting surfaces must be machined to high flatness and parallelism (typically better than 0.05 mm) to avoid introducing preloads or bending moments that bias the zero point [9]. The stiffness and design of the mechanical interface (adapter plates, tool changers) between the sensor and the end-effector also affect force transmission and must be considered part of the overall measurement chain. In harsh environments, exposure to dust, moisture, coolants, or corrosive chemicals can compromise sensor integrity. While many industrial sensors are rated to IP65 or IP67 standards for ingress protection, prolonged exposure or high-pressure washdowns may require specialized seals or hermetic packaging [10]. In surgical applications, sensors must withstand repeated sterilization cycles using autoclaves (high-temperature steam) or chemical agents like ethylene oxide, which can degrade adhesives, seals, and electronic components over time [11].

Application-Specific Limitations and Trade-offs

The utility of force-torque sensors is bounded by inherent physical and practical limitations that vary by application. In robot-assisted surgery (RAS), an obvious drawback is the loss of haptic sensation, depriving surgeons of a natural source of information about interaction forces [12]. This limitation is compounded by the challenge of miniaturization. Developing sensors small enough to fit within the constrained diameter of laparoscopic instruments (often less than 10 mm) while maintaining adequate sensitivity, range, and sterilization compatibility remains a significant engineering hurdle [13]. Consequently, the lack of tactile feedback has remained an important limitation of robotic-assisted thoracic surgery (RATS) and other minimally invasive procedures, driving research into alternative sensing modalities like instrument vibration analysis or tissue deformation modeling [14]. For dynamic applications such as robotic assembly, polishing, or physical human-robot interaction, sensor bandwidth is a critical but double-edged parameter. While a high bandwidth is necessary to capture rapid transients for stable force control, it also makes the sensor more susceptible to high-frequency noise and vibrations from the robot's own motors and gearboxes [15]. Designers must often implement low-pass filters, but these introduce phase lag that can destabilize the control loop if not properly accounted for [16]. This creates a fundamental trade-off between measurement fidelity and control stability. Finally, the economic and lifecycle costs must be considered. High-precision, multi-axis sensors represent a significant capital investment. Beyond the initial purchase, costs include integration engineering, calibration equipment or services, and potential downtime for maintenance or replacement [17]. The total cost of ownership must be justified by the value added through improved quality, reduced scrap, enhanced safety, or new capabilities like adaptive machining or precise assembly [18].