Multi

Industrial robots are more affordable than ever for precision assembly and high-speed picking/packing tasks. With improvements to capabilities like vision, each new generation delivers more human-like dexterity and flexibility. A reliable and cost-effective sense of touch now lets them handle fragile objects to fulfill an even wider variety of tasks and interact more safely with humans.

Several techniques have been explored to introduce tactile sensing for robots. These include liquid metal sensors that measure the resistance of a liquid metal flowing in microfluidic channels, which is modulated by external forces. Although this type of sensor can be incorporated all around the robot fingertips, it does not measure localized 3D force. Instead, the distributed forces are mapped to a resistance change. A group of such sensors can detect a pattern. Hence this technique is mostly used in specialized classification tasks after dedicated training.

Another approach uses a high-end optical camera to measure the deformation of an elastomeric material embedded in, or covering, the gripping surfaces. This technology is commercially available and already used in smart multi-modal robotic grippers. However, the camera needs a substantial pixel array, and transmitting the data at a video rate for analysis demands significant communication bandwidth and power.

A different optical-based solution overcomes some of these challenges by using a quadrant of photodiode detectors instead of a full camera. A light is shown into an elastic dome from the inside and the detectors sense the deformation of the dome due to the contact forces. However, the power consumption is several times larger than that of typical 3D magnetometers, which can provide an even simpler and more efficient alternative.

In such magnetic sensors, a magnet is embedded in an elastomeric material similar to that used in the camera-based system. A rear-mounted magnetometer provides 3D force sensing by measuring the change in magnetic field caused by displacement of the magnet as the elastomer is deformed. Several such sensors have been demonstrated using a single-output magnetometer, which can be imagined as a tactile pixel, or "taxel". Researchers have built various configurations, from simple single-pixel and 2×2 arrays to a 15 mm2 continuous magnetic skin consisting of magnetic microparticles. Single-pixel magnetic force sensors like these have achieved sub-1 mm resolution by combining the sensor with sinusoidal magnetization of a flexible film and deep-learning techniques.

The magnetic sensors referenced here have used the Melexis MLX90393 single-pixel magnetometer. While the benefits of magnetic sensing include relatively low power and minimal computing and communication overhead, single-pixel sensing is vulnerable to interference from external magnetic fields. The magnetometer output can be distorted by unrelated effects in the vicinity, such as activating an electric motor, the presence of other magnets, or variations in the earth's magnetic field.

A magnetic force sensor with multiple nearby pixels inside the same integrated circuit (IC) package (Figure 1) can provide immunity to stray fields by enabling differential measurement. This article describes how the Tactaxis gradiometric multi-pixel magnetic sensor was built and tested.

The Melexis MLX90372 linear-displacement sensor provides a convenient platform to demonstrate the gradiometric sensing principle. This sensor usually outputs the angular displacement along an arc. However, configuring the device in test mode allows direct access to the raw magnetic readouts of the individual pixels from the memory. The sensor is housed in a standard 5 mm × 4.3 mm × 0.9 mm TSSOP package and contains two side-by-side CMOS dies, with two pixels per die. Hence, this single, compact component contains four magnetic pixels placed about 2 mm apart, which enables the gradient of the magnetic field to be measured. Each pixel senses the normal component of the field Bz and the Bx in-plane component.

On top of the IC package, there is a soft elastomer that contains an embedded disk magnet with axial magnetization. Using a cylindrical-shaped sample of elastomer minimizes the magnet tilt and presents a compliant surface. Applying contact force to the elastomer displaces the magnet, which modulates the magnetic field pattern. The four magnetic pixels each sense this displacement and so can detect the effects of normal and lateral forces. For a normal displacement of the magnet, the gradiometric component ∂Bx/∂x is the most impacted. Conversely, for a lateral displacement, the impact is mostly on the gradiometric component ∂Bz/∂x.

Note that the sensor measures only displacement of the magnet. The connection between displacement and applied force depends on other factors and requires further calibration and calculation. The magnet size, elastomer hardness, and elastomer cylinder diameter all influence the magnitude of the sensor output when a force is applied. A larger, stronger magnet increases the signal-to-noise ratio (SNR), with little impact on the full-scale force. A harder elastomer enables a larger full-scale force. However, for the same force applied, the displacement of the magnet will be smaller compared to a softer material. Hence increasing the hardness also reduces the change of magnetic signal, hence the SNR. Finally, the elastomer diameter acts as a scaling factor between the overall force and the localized pressure just above the magnet. A larger diameter distributes the force over a larger area thereby lowering the SNR, while accommodating larger full-scale force.

Figure 2 shows the functional block diagram of the signal chain.

The signal processing is performed offchip and begins by scaling the eight digital output signals of the chip to correct for the sensitivity drop of the Hall effect with increasing temperature (−0.5 %/°C).

The stray magnetic fields are then rejected by using combinations of field components. The mean of Bx field and Bz field are first removed, leaving remaining terms that are related to the gradient of the magnetic field. Effectively, the force sensor algorithm processes the magnetic field differences within the two dies.

The feature augmentation block calculates the norm √(Bx2 + Bz2) in each sensing pixel, producing a 12-dimensional vector signal {Bx, Bz, Bnorm} at each pixel.

The last step generates a new vector containing all second-order polynomial combinations of the 12-dimensional vector, including interaction terms. This produces a vector signal of dimension 91.

Finally, the inference stage calculates the force and planar-torque values from the vector signal using a 91-by-5 weight matrix. The weights are obtained through a training procedure using a reference load cell mounted on a 3-axis moving platform to stress the elastomer by applying a known displacement. The force from the load cell and corresponding magnetic signals from the sensor were measured and saved at 13,000 positions throughout a displacement range of 1.5 mm depth and 1.1 mm radius.

To demonstrate how the effects of external fields are eliminated, the sensor was placed between two Helmholtz coils generating ±2 mT (Figure 3a). A field of equivalent strength can be experienced at about 3 cm distance from common home appliances.

The force measured by the sensor using the gradiometric concept is shown in Figure 3b (blue curve), demonstrating that the stray-field error is limited to 0.3% of full scale. The prototype sensor was then reconfigured to operate as a plain magnetometer without stray-field rejection, emulating the behavior of the earlier single-pixel sensors. The stray field leaks directly into the signal path with no rejection, producing errors of up to 20 % at −2 mT (red curve). This is almost two orders of magnitude larger than the gradiometric sensor.

The prototype Tactaxis sensor was mounted on a commercial robotic hand. A basic force-control algorithm was implemented to have the hand gently grasp a balloon. Figure 4 shows the demonstration setup.

Using the sensor as a single-pixel plain magnetometer, the force is initially well regulated in the absence of stray-field disturbance. Introducing a stray field using a magnet corrupted the force sensor, causing the hand either to release or crush the balloon depending on the polarity.

When using the Tactaxis sensor in its proper multi-pixel mode, the force remained properly regulated at all times, unaffected by the approaching magnet down to a distance of a few centimeters.

Table 1 compares the properties of the Tactaxis multi-pixel sensor with commercially available single-pixel magnetic, optical, and piezo-electric force sensors.

The single-pixel magnetic sensor is compact and achieves 3D force vector sensing with state-of-the-art resolution, particularly in multi-sensor configurations. However, the sensitivity to stray fields remains a key limitation.

The optical sensor is naturally fully immune to magnetic stray fields and offers similar 3D force-sensing performance. Although an excellent functional fit for robotic hand integration, the discrete optical components drive the cost upwards.

Piezoresistive sensors have the advantage of a small form factor, about the size of an IC package, and achieve competitive force resolution but can only sense the normal force.

In contrast, the prototype Tactaxis multi-pixel sensor delivers the known assets of magnetic sensors, namely 3D-force sensing, softness, economy, and compactness, with superior immunity to real-world parasitic stray fields. Hence this gradiometric sensing concept has advanced the robustness of force sensing for robotic applications.

This article was written by Gael Close, Global Innovation Manager, Melexis. For more information, contact Mr. Close at This email address is being protected from spambots. You need JavaScript enabled to view it..

This article first appeared in the May, 2023 issue of Sensor Technology Magazine.

Blog