Time-of-flight signal processing for FTIR-based tactile sensors

Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38909 Time-of-flight signal processing for FTIR-based tactile sensors JORGE GARCIA-PUEYO,1,†,* SERGIO CARTIEL,1,†EMMANUEL BACHER,2MARTIN LAURENZIS,2AND ADOLFO MUÑOZ1 1Universidad de Zaragoza, I3A, Zaragoza, Spain 2French-German Research Institute of Saint-Louis, 68301 Saint-Louis, France †These authors contributed equally to this work. *[email protected] Abstract: Optical tactile sensors offer a promising avenue for advanced sensing and perception. We focus on frustrated total internal reflection (FTIR) tactile sensors that utilize time-of-flight (ToF) measurements. We analyze the complex behavior of ToF signals within optical waveguides in the time domain, where phenomena like internal reflections and scattering significantly influence light propagation, especially in the presence of touch. Leveraging this analysis, we develop a real-time processing algorithm that enhances FTIR tactile sensing capabilities, allowing for precise detection. We evaluate our algorithm on an OptoSkin sensor setup, demonstrating a significant improvement in multi-touch detection and contact shape reconstruction accuracy. This work represents a significant step towards high-resolution, low-cost optical tactile sensors, and advances the understanding of time-resolved light transport within waveguides and in scattering media. © 2025 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement 1. Introduction Tactile and pressure sensing are essential for interacting with the world and are present in most living beings. As such, this sensing modality has been developed artificially, for applications such as robotics, and there has been a proliferation of tactile sensors with different sensing principles and sizes. The general adoption of tactile sensors is subject to challenges such as robustness, fabrication cost, deployment methodology and low-latency processing of the sensors. Large-area sensors are even more challenging due to the size of their coverage. Most of the large area tactile sensors are array-based, being composed of smaller sensor units (taxels) arranged in a surface, with cost proportional to surface area. In this regard, different technologies have been used, including piezo-resistive [1], capacitive [2], magnetic [3] or optical [4]. Among optical approaches for touch sensing, we can differentiate between those based on vision of surface deformation (TacTip [5], GelSight [6], Digit [7]) and those based on light transmission, which analyse the light changes due to contact but do not need the light detector to directly observe the contact point, lifting limitations on sensor design and setup. Light transmission sensors are based on diverse sensing principles like frustrated total internal reflection (FTIR) [8,9], liquid lenses [10] or optical fibers [11,12]. Recent works in tactile sensors based on light transmission have presented new approaches using optical time-of-flight (ToF) sensors together with optical waveguides for detecting touch [13,14]. In these setups, ToF sensors do not need to cover the whole surface area but only its perimeter, which, in practice, reduces fabrication time, cost and deployment time. Of particular interest is the recent ToF FTIR-based tactile sensor, OptoSkin [14], that uses multiple ToF sensors on the perimeter of a planar optical waveguide. Light emitted by the ToF sensors propagates within the waveguide by total internal reflection (TIR). When an object makes contact with the waveguide, TIR is disrupted, causing light to scatter and escape. This #570548 https://doi.org/10.1364/OE.570548 Journal © 2025 Received 11 Jun 2025; revised 23 Jul 2025; accepted 24 Jul 2025; published 4 Sep 2025 Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38910 change in light behavior is captured by the detector element of the ToF sensor and enables the reconstruction of the touch position and pressure. Still, the original work where the OptoSkin sensor is presented overlooks most of the ToF signal structure within waveguides, and therefore demonstrates detection of just one or two simultaneous touches. In this work, we devise a physically-based model to describe the structure of ToF signals inside waveguides, accounting for most optical phenomena, that becomes an analysis tool for ToF FTIR sensors. We then invert this model, and propose a real-time algorithm that processes the signal of ToF FTIR tactile sensors to greatly enhance contact and pressure detection. We evaluate its performance on a setup based on the recent OptoSkin sensor [14], demonstrating that our processing algorithm enhances the detection of simultaneous multiple contact points and improves reconstruction of contact area shapes. 2. Forward model: light propagation within optical waveguides Large-area tactile sensors based on FTIR employ light emitters and detectors together with optical waveguides to direct light from one point to another by TIR. When something touches the waveguide, it produces FTIR and light is scattered. Light propagation within waveguides is subject to several optical phenomena that lead to signal degradation. This becomes particularly problematic for time-resolved signals, such as those obtained from ToF sensors. As a result, ToF FTIR sensors experience limitations when detecting touch and pressure with raw unprocessed signals, which we overcome in this paper by applying our algorithm to process them. In this section we give an overview of the optical phenomena of light propagation within waveguides, relating it to the LiDAR (Light Detection and Ranging) equation. 2.1. Time-of-flight sensors and the LiDAR equation ToF sensors determine the distance between the sensor and an object by measuring the travel time of pulsed light. They are composed of a light emitter (e.g., a laser) and a detector (e.g., single-photon avalanche diodes, SPAD). The detector outputs histograms of photon counts over time (Fig. 1(a)). The histogram bin containing the highest peak of light is assumed to represent direct light reflecting from the object. The time value t peak associated to such bin is used to compute the distance d = t peak c / 2 between the sensor and the object, being c[in m / s] the speed of light in air and the factor 1/2 to account for light traveling forth and back. Light measurements by ToF sensors can be described by the LiDAR equation [15], which quantifies the received photon counts (or light power) by the detector Q(t)[in Watts] as Q(t)=Pe(t)AϵG(t)β(t)T(t), (1) where P e( t ) [in Watts] is the emitted photons (or emitted power) by the laser, A[in m 2 ] is the aperture area of the detector, ϵ is the dimensionless system efficiency. The remaining terms describe the interaction of light with the medium: the geometric factor G ( t ) [in m −2 ] represents light divergence as it propagates, the transmittance term T ( t ) [dimensionless] represents the attenuation of light due to absorption and scattering properties of the medium, and the backscattering coefficient β( t ) [dimensionless] quantifies the amount of light scattered back towards the detector from the medium at a specific range. For hard surfaces, β( t ) relates to the object’s albedo at its distance and is zero beyond it. 2.2. Our forward model When ToF sensors are coupled to waveguides, the temporal profile of light signals is different due to multiple light propagation effects (Fig. 1(b)-(e)). Therefore, detection of light peaks in such signals is suboptimal and prone to significant errors. To address this, we propose to define a forward model that describes the detected light signal from ToF sensors coupled to waveguides Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38911 Fig. 1. Optical phenomena affecting light propagation in waveguides (up row) and the corresponding time-of-flight (ToF) signal (bottom row). (a) The ToF sensor emits a pulse of light that propagates through air, determining the object distance from the time-of-flight of the reflected light, which is easily distinguishable. (b) When coupled to a smooth dielectric waveguide, light travels efficiently through total internal reflection (TIR), allowing the ToF sensor to detect the waveguide’s end accurately, which appears later due to the higher index of refraction of the waveguide. (c) If the waveguide’s surface is rough, for instance due to fabrication imperfections, it produces some backscattering, increasing the signal at earlier times. (d) The presence of scattering particles (green dots) further disrupts light propagation, increasing backscattering and, thus, the signal complexity. (e) When there is a contact with the waveguide, it disrupts TIR, a phenomenon known as frustrated total internal reflection (FTIR), redirecting some light back to the sensor (blue path). Our forward model (Section 2) accounts for all these effects. We invert it (Section 3) to devise a signal processing algorithm that enables to identify the contact signal with respect to the rest of the optical phenomena signal. and use it, in Section 3, to process the ToF signal for optimal detection of peaks of light associated to touch or pressure. Unlike typical LiDAR applications that assume a single backscattering source (e.g., the atmosphere in the remote sensing field), our scenario involves multiple elements that can generate backscattering, like the waveguide’s interface, waveguide’s material and the object contacting the waveguide. Therefore, our proposed forward model extends the LiDAR equation [15] to include these multiple backscattering sources as Pr(t)=Qint(t)+Qvol(t)+Qtouch(t)+Nms(t)+Na, (2) where Q int( t ) ,Q vol( t ) and Q touch( t ) describe the power received due to backscattering interactions with the interface of the waveguide, the material composing the volume of the waveguide and the object touching the waveguide, respectively. Additionally, we include N ms( t ) [in Watts] to account for received light due to multiple scattering and N a [in Watts] for the ambient light (Section 2.6). We decompose each component Qias Qi(t)=KG(t)βi(t)T(t), (3) following the LiDAR equation (Eq. (1)), where K[in Watts · m 2 ] summarizes the performance of the system depending on the emitted power P e( t ) , the area of the detector Aand system efficiency ϵ, since they depend on the characteristics of the ToF sensor. In the following sections, we describe in detail the geometry factor G ( t ) (Section 2.4), the transmittance term T ( t ) (Section 2.5) and the backscatter coefficients βi( t ) for each backscattering source: interface of the waveguide βint( t ) (Section 2.3), the volume of the waveguide βvol( t ) (Section 2.5) or the object touching the waveguide βtouch(t)(Section 2.7). The term Kis common to all the components Q i( t ) . Specifically, Q touch( t )= 0 when there is not contact with the waveguide. While the contribution from Q int( t ) ,Q vol( t ) and N ms( t ) depend Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38912 on the optical characteristics of the setup, which will remain stable, Q touch( t ) reflects contact and pressure on the waveguide. Our goal is to detect variations of the time-of-flight measurements to obtain βtouch(t), the touch backscattering, which we consider directly correlated with contact. 2.3. Dielectric waveguides An optical waveguide consists on a dielectric slab, where the core region has a higher refractive index n 1 than both the medium above the core n 0 and the deposited substrate n 2 (n 1> n 2≥ n 0) . If the medium above the core is air, n 0≈ 1. These different refractive indices cause TIR, allowing light to propagate inside the waveguide due to light bouncing at both interfaces with an incident angle greater than the critical angle. When the top and bottom boundaries of the waveguide are perfectly smooth and parallel, light travels by bouncing in a zig-zag pattern (Fig. 1(b)). Furthermore, the amount of backscattering produced by the interface would be negligible, Q int( t ) ∝ βint( t ) ≈ 0, except for the signal received due to the bounce in the end of the waveguide. However, fabrication processes might introduce micro-scale geometric irregularities in the interfaces, breaking the assumption of being smooth. Rough interfaces might introduce light leakage outside the waveguide and increase the amount of backscattering βint( t )> 0 so its effect, while probably still small, would not be negligible (Fig. 1(c)). Additionally, surface roughness also increases multiple scattering Nms(t). 2.4. Geometric factor Due to light divergence as it propagates, there is a light intensity falloff that is proportional to the inverse squared distance. This phenomena is accounted for in the geometric factor G(t)=1 (c1t)2, (4) where c 1= c / n 1 is the speed of light inside the dielectric waveguide and n 1 is the refractive index of the dielectric waveguide. 2.5. Scattering and extinction Optical waveguides are composed of materials that may act as participating media, which differentiably interact with light, resulting in signal loss and aggravating the problem of nonstraight light propagation (Fig. 1(d)). We characterize optical waveguides by analyzing their optical parameters. Specifically, we use the absorption coefficient µa , which describes how much light is absorbed by the volume, and the scattering coefficient µs , which describes how much light interacts with the media, both in m −1 . We compute the extinction coefficient, µt=µa+µs [in m −1 ] which describes the loss of intensity as light traverses a medium both from absorption and scattering. The extinction coefficient affects the power received by the detector through the transmittance term T(t)=e−2µt(c1t), (5) making the signal received by the detector lower, and where the factor 2 stands for the forth and back paths. The scattering coefficient µs also affects the light propagation direction. This might produce backscattering, making light come back to the sensor and increasing the amount of light arriving to it at every timestep, Q vol( t ) ∝ βvol( t ) ∝ µs , and slightly blurring the signal. Additionally, it will partially increase the overall multiple scattering Nms(t). Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38913 2.6. Ambient lighting In addition, the detected signal will also include ambient light N a , which comes from different sources such as sunlight and artificial lighting. Consequently, ambient light must be removed before evaluating the detected signal. In ToF sensors, the ambient signal at every timestep is estimated by collecting the light during the instant before the light pulse emission. Additionally, ToF sensors are usually equipped with optical filters that block light coming from wavelengths different from the wavelength of the emitter. In any case, ambient light is usually considered constant for a single histogram measurement, hence its independence from time in the forward model. 2.7. Frustrated total internal reflection In optical waveguides, FTIR occurs when the outside medium (usually air) is replaced with another medium with a larger index of refraction. This medium interferes with TIR and causes light to exit the waveguide. When this new medium is highly scattering or even opaque, light scatters in all directions, reentering the waveguide and arriving to the detector (Fig. 1(e)). In FTIR-based sensors, this outside medium is a contact point, from which touch is detected. This effect is increased when the surface of the waveguide is deformable, so that pressure alters the surface orientation and hence the incoming light angle, which might be now below the critical angle. In our model, touch is represented by the backscatter coefficient βtouch( t )> 0, which is in practice the value we want to obtain from the raw signal Pr(t). 3. Inverse model: time-resolved signal correction ToF sensors are able to detect objects in air because the associated light peak is easily distinguishable (Fig. 1(a)). However, as discussed in Section 2, the light signal structure changes drastically when ToF sensors are coupled to a waveguide. In this case, the detector of the ToF sensor collects light from multiple sources, making the default histogram-based distance estimation algorithms ineffective. In this section, we propose a signal correction algorithm to process the raw captured signal P r( t ) (Fig. 2(b)) and isolate the signal coming from touch to estimate βtouch( t ) (Fig. 2(e)). 3.1. Baseline removal ToF sensors cannot distinguish the source of the light they are detecting, so we have to isolate the signal received due to objects touching the waveguide. Initially, we perform a calibration step that consists on measuring the signal detected by the ToF sensor when no object touches the surface of the waveguide. We named this baseline signal B ( t ) and it accounts for signal coming from ambient light N a , from single backscattering after interacting with the interface of the waveguide Q int( t ) (including the end of the waveguide that creates the final peak in Fig. 1(b)-(e)), from single backscattering after interacting with participating media of the volume of the waveguide Q vol( t ) and from multiple scattering N ms( t ) of multiple bounces light paths. The resulting signal after baseline removal ˆ Qtouch(t)(Fig. 2(c)), computed as ˆ Qtouch(t)=Pr(t)−B(t) ≈Pr(t)−(Qint(t)+Qvol(t)+Nms(t)+Na),(6) is an approximation of the signal received due to objects touching the waveguide. While each of the components (i.e. Q int( t ) ,Q vol( t ) ,N ms( t ) and N a ) of the baseline signal B ( t ) could be analytically modeled, these models would probably be too ideal compared to the realistic imperfections of non-uniform materials or fabrication irregularities. Therefore, a measured baseline is preferred. Note that a portion of ambient light N a is already removed due to physical filtering previous to baseline subtraction. Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38914 Fig. 2. ToF signals from a sensor coupled to a waveguide during two touch interactions. Each column corresponds to a different signal processing method, and each row shows a different contact location: close (top) and far (bottom) from the ToF sensor. (a) Diagram of the close and far contact points on the waveguide. (b) Raw: Unprocessed ToF measurements P r( t ) . The blue dotted lines show baseline signals (no touch), and the orange lines show signals during touch. For the close contact point, distinct peaks are observed corresponding to light entering the waveguide, contact interaction and waveguide end. For the far contact point, only the light entering the waveguide and the waveguide end peaks are visible. (c) Baseline removal: baseline signal B ( t ) is subtracted from touch signal P r( t ) , revealing the peaks at contact locations. However, the intensity of the contact peaks vary significantly between the close one (clearly visible) and the far one (slightly visible). (d) Decay correction: the geometric factor G ( t ) and extinction term T ( t ) are compensated to normalize signal intensity over time. The close contact point becomes clearer but the far contact point remains suppressed. (e) Ours: combines baseline removal and decay correction, yielding consistent touch signal profiles ˆ βtouch(t)regardless of contact location. 3.2. Decay correction We can assume that when ˆ Qtouch( t ) ≈ 0, nothing is touching the waveguide, and when ˆ Qtouch( t )> 0, something is touching it. However, light presents an intensity decay dependent on the distance traveled by the light due to the geometry factor G ( t ) and the transmittance term T ( t ) (Sections 2.4 and 2.5). As a consequence, objects closer to the sensor will appear brighter than the ones further, as seen when applying only baseline removal in Fig. 2(c). To ensure that touches at different distances produce signals of comparable magnitudes, it is necessary to account for the light decay, that depends on the distance traveled by light, as ˆ βtouch(t)= ˆ Qtouch(t) G(t)T(t)=ˆ Qtouch(t)(c1t)2 e−2µt(c1t). (7) This approach compensates for the light intensity falloff observed in the raw signal P r( t ) (Fig. 2(b)) and, when applied on the signal after baseline removal ˆ Qtouch( t ) , the processed signal ˆ βtouch( t ) (Fig. 2(e)) is sufficiently consistent for automatic extraction of touch information. Note that just applying decay correction on the raw signal is not enough since the touch signal is still mixed with the baseline signal and far contact points are not clearly visible (Fig. 2(d), bottom row). 3.3. Implementation We describe the implementation of our signal correction processing in Algorithm 1. The inputs are the raw captured signal P r( t ) , a previously acquired baseline signal B ( t ) and known parameters of the optical setup (bin temporal resolution ∆ tof the ToF histogram, refractive index n 1 , absorption coefficient µa and scattering soefficient µs of the waveguide’s material). We first perform baseline removal (Section 3.1) to isolate the touch signal ˆ Qtouch( t ) , followed by decay Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38915 correction (Section 3.2) to normalize signal intensity across different distances. A more detailed analysis of the theoretical and experimental time complexity of our algorithm is provided in Supplement 1, Section 1. Algorithm 1. Our time-resolved signal correction 4. OptoSkin and tactile maps Our analysis and modeling of light propagation within waveguides (Section 2), along with the proposed signal correction algorithm (Section 3), are general and broadly applicable to any light sensor embedded in a volumetric medium. While this paper focuses on optical tactile sensors based on TIR or FTIR [4,8,9,13,16–19], the underlying approach can also be extended to other domains where light transport in scattering or refractive media plays a critical role, for instance underwater imaging [20–22] or detection through fog [23]. We choose to assess its performance (Section 5) on the ToF FTIR-based OptoSkin sensor [14]. In this section, we describe our experimental OptoSkin setup and the process to generate tactile maps to represent touch on the waveguide surface. 4.1. Experimental setup We build our OptoSkin setup following the original work [14], strategically placing ToF sensors along the edges of the waveguide. As shown in Fig. 3, for the experiments, we placed six ToF sensors around a square waveguide with a size of 300mm × 300mm and a thickness of 5mm. The waveguide consists of a commercial two-component transparent soft silicone material (Troll Factory, Type 19, TFC4190-T19 [24]) with the following optical characteristics: refractive index of n 1= 1.4, scattering coefficient of µs= 0.6cm −1 , and absorption coefficient of µa= 0.14cm −1 . See Supplement 1, Section 3 for more information on the waveguide’s silicone material preparation and physical properties. The positions of the sensors were chosen to ensure a comprehensive coverage of the tactile area by at least one sensor. We utilized the TMF8828 ToF sensor from AMS-OSRAM [25], which integrates a low-power pulsed VCSEL (vertical cavity surface-emitting laser diode) and SPAD technology within a single module. Additionally, it features an electronic control unit enabling the configuration of the sensing zones of the SPAD array (binning) and the processing and formatting of data (time-of-flight histograms) to be transmitted to an external device via an I2C bus. The binning on sensor level is used to reduce the overall data acquisition, processing and transmission time. The key optical parameters of the TMF8828 ToF sensor are the bin temporal resolution, the field of illumination (FoI) and field of view (FoV). The bin temporal resolution is ∆ t ≈ 100ps, Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38916 Fig. 3. Sensor layout (a) and image (b) of our OptoSkin setup used in the performance experiments (Section 5). Six ToF sensors (S 0 , . . . ,S 5 ) are strategically placed around a 300mm × 300mm waveguide to ensure a comprehensive sensor coverage. The waveguide consists of a 5mm thick soft silicone rubber waveguide layer molded onto an aluminum substrate. which is equivalent to a distance resolution (in air) of ∆ d =∆ t c 2= 1.5cm. When coupled to a waveguide, the equivalent distance resolution has to take into account the refractive index as ∆ d ′=∆ t c1 2=∆ t c 2n1≈ 1.1cm. The illumination (FoI) is fixed to 41 ◦× 47 ◦ , the FoV can be adjusted by selection of the sensing zone configuration. In general, standard configurations of 3 × 3, 4 × 4, 3 × 6, and 8 × 8 detection areas can be selected. Further, this sensor features the option to customize the SPAD binning to user defined zone configuration. In our setup, we defined a mode with 7 × 1 sensing areas. This mode reduces the data acquisition and preprocessing load while providing a maximum horizontal resolution within an FoV of 41 ◦× 52 ◦ in air. Due to optical refraction, in the waveguide material, the FoI and FoV is reduced to 29 ◦× 33.5 ◦ and 29◦×37◦, respectively. A custom electronic board built around an ESP32 microcontroller is used to set the ToF sensor measurement parameters, read data, and synchronize the sensors. This synchronization prevents sensors from interfering with each other when observing the same region of the tactile surface and allows the I2C bus to operate at the maximum read frequency. The number of sensors on the I2C bus, the measurement zones, and laser repetitions mainly influence our data acquisition frequency. For example, with a typical standard configuration of four 4 × 4 sensing areas, the data acquisition rate is 4.5 samples per second, with each ToF sensor recording 250,000 samples. In the optimized 7 × 1 mode, the acquisition rate increases by a factor of two to 9 samples per second. 4.2. Tactile map creation Touch location reconstruction is based on creating a touch likelihood heatmap (a tactile map) by projecting the captured time-of-flight histograms on the surface of the waveguide, considering the time-of-flight of the light and the FoI/FoV of the ToF sensor. For that, we transform the time-of-flight histograms into distance measurements by accounting for the round trip and the speed of light, so distance d = t c1 2= t c 2n1 where cand c 1 are the speed of light in air and inside the waveguide respectively, and n1=1.4 is the waveguide refractive index. Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38917 The tactile map C( x,y ) of a ToF sensor in position ( x s ,y s) and whose FoV covers the angles θ0 to θ1(measured with respect to the y-axis) can be reconstructed as θ=atan2(x−xs,y−ys) C(x,y)=⎧ ⎪ ⎪⎨ ⎪ ⎪ ⎩ H(︃2n1√(x−xs)2+(y−ys)2 c0n0)︃if θ∈ [θ0,θ1] 0 otherwise (8) where Hrepresents the temporal histogram, either raw (H ( t )= P r( t ) ), with only baseline removal (H(t)=Pr(t)−B(t)) or with our full processing (H(t)=ˆ βtouch(t)). Each tactile map Ci( x,y ) is created from the histogram H i( t ) corresponding to ToF sensor S i and assumes that the waveguide is planar or can be approximated to a planar surface (since the thickness of the waveguide is much smaller than its width and height dimensions). Reconstructions involving a setup with multiple ToF sensors around the waveguide can be created by adding the tactile maps Ci( x,y ) from each individual ToF sensor. For that, each individual tactile map must be rotated according to the looking direction of each individual ToF sensor. See Supplement 1, Section 2 for a pseudocode algorithm of the tactile map creation. In the original OptoSkin work, the signal processing for tactile map reconstruction was limited to removing the baseline (Section 3.1), allowing to detect a maximum of two simultaneous touches in the surface of the optical waveguide. In this work, we show how to improve tactile map reconstructions by processing the ToF signal using our proposed algorithm (Section 3), increasing the accuracy of the OptoSkin sensor reconstructions, especially when detecting multiple simultaneous touches and shapes. 5. Performance evaluation In this section, we evaluate the sensing performance, analyzing both the effect of physical filtering and various signal processing algorithms with our experimental OptoSkin setup in real-world experiments. Particular emphasis is placed on the advantages of our signal processing algorithm, which significantly enhances the sensor’s ability to accurately detect and interpret multi-touch and complex shapes inputs. Signal processing is analyzed in three key interaction scenarios: single-touch detection, simultaneous multi-touch detection, and shape reconstruction. All performance evaluation tests were carried out in a room inside a building under standard conditions (approx. 20 ◦ C and 40% humidity) and under homogeneous artificial lighting (halogen tubes, 500 lux). Contact interactions involved pressures of approximately 300g / cm 2 , ensuring sufficient contact with the waveguide’s surface to produce FTIR in our OptoSkin setup. 5.1. Ambient light physical filtering Before analyzing the effect of our signal processing algorithm, we explore the limits of the physical filtering provided by the optical filters of our ToF sensors. For this purpose, we devise a setup involving a 3D printed waveguide, as shown in Fig. 4(a). For the measurements, the entire surface of the waveguide was illuminated with a laboratory white light source consisting of a halogen lamp and a stabilized power supply positioned perpendicular to it. By adjusting the electrical current, we were able to vary the irradiance. In order to address different spectral ranges, visible ( < 800 nm) and infrared ( > 800 nm), optical filters (band pass filter, BPF) were placed in the light path. The contact point on the waveguide was emulated by sticking an adhesive tape to the back of the sensor skin. This allowed us to investigate how the ToF sensor responds to FTIR due to the tape in the presence of varying ambient light. The baseline signal was recorded without ambient light. Figure 4(b) shows the contact signal (solid lines) from the adhesive tape for different levels of irradiance illumination in the visible range as raw histograms (up) and as processed by our Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38924 References 1. W. Zheng, K. Liu, D. Guo, et al., “A large-area tactile sensor for distributed force sensing using highly sensitive piezoresistive sponge,” in International Conference on Robotics and Automation (IEEE, 2024), pp. 1063–1069. 2. F. Giovinazzo, F. Grella, M. Sartore, et al., “From cyskin to proxyskin: Design, implementation and testing of a multi-modal robotic skin for human-robot interaction,” Sensors 24(4), 1334 (2024). 3. H. Hu, C. Zhang, X. Lai, et al., “Large-area magnetic skin for multi-point and multi-scale tactile sensing with super-resolution,” npj Flexible Electron. 8(1), 42 (2024). 4. T. Pereira, S. Rusinkiewicz, and W. Matusik, “Computational light routing: 3d printed optical fibers for sensing and display,” ACM Trans. Graph. 33(3), 1–13 (2014). 5. B. Ward-Cherrier, N. Pestell, L. Cramphorn, et al., “The tactip family: Soft optical tactile sensors with 3d-printed biomimetic morphologies,” Soft Rob. 5(2), 216–227 (2018). 6. R. Li, R. Platt, W. Yuan, et al., “Localization and manipulation of small parts using gelsight tactile sensing,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (2014), pp. 3988–3993. 7. M. Lambeta, G. Kammerer, D. Jayaraman, et al., “Digit: A novel design for a low-cost compact high-resolution tactile sensor with application to in-hand manipulation,” IEEE Robotics and Automation Letters PP, 1 (2020). 8. J. Y. Han, “Low-cost multi-touch sensing through frustrated total internal reflection,” in Proceedings of the 18th Annual ACM Symposium on User Interface Software and Technology (Association for Computing Machinery, 2005), pp. 115–118. 9. R. Wattanaparinton and K. Takemura, “Vision-based tactile sensing using multiple contact images generated by re-propagated frustrated total internal reflections,” in International Conference on Systems, Man, and Cybernetics (IEEE, 2022), pp. 962–967. 10. R. Hu, H. Yang, L. Wang, et al., “Flexible optical tactile sensor based on a liquid-membrane lens structure,” Appl. Opt. 62(26), 6952–6960 (2023). 11. T. Xu, L. Li, Y. Wang, et al., “Highly sensitive soft optical fiber tactile sensor,” Opt. Express 30 (19), 34064–34076 (2022). 12. L. Li, F. Yang, Q. Ma, et al., “Integrating an optical fiber sensor and artificial intelligence for enhanced tactile sensing in human-computer interaction,” Optica 12(2), 263–273 (2025). 13. J.-T. Lin, C. A. Newquist, and C. K. Harnett, “Multitouch pressure sensing with soft optical time-of-flight sensors,” IEEE Trans. Instrum. Meas. 71, 1–8 (2022). 14. E. Bacher, S. Cartiel, J. García-Pueyo, et al., “Optoskin: Novel lidar touch sensors for detection of touch and pressure within wave guides,” IEEE Sens. J. 24(20), 33268–33280 (2024). 15. U. Wandinger, Introduction to Lidar (Springer, 2005), pp. 1–18. 16. J.-H. Lee and C.-H. Won, “High-resolution tactile imaging sensor using total internal reflection and nonrigid pattern matching algorithm,” IEEE Sens. J. 11(9), 2084–2093 (2011). 17. A. Lavatelli, A. Zanoni, E. Zappa, et al., “On the design of force sensors based on frustrated total internal reflection,” IEEE Trans. Instrum. Meas. 68(10), 4065–4074 (2019). 18. H. Krauss and K. Takemura, “Stretchable optical waveguide sensor capable of two-degree-of-freedom strain sensing mediated by a semidivided optical core,” IEEE/ASME Transactions on Mechatronics 27(4), 2151–2157 (2022). 19. J. Hu, D. Cao, Y. Li, et al., “Polymer-based optical waveguide tactile sensing method for 3-d surfaces,” IEEE Sens. J. 23(8), 8761–8768 (2023). 20. A. Maccarone, K. Drummond, A. McCarthy, et al., “Submerged single-photon lidar imaging sensor used for real-time 3d scene reconstruction in scattering underwater environments,” Opt. Express 31(10), 16690–16708 (2023). 21. M. Shangguan, Y. Li, Y. Mo, et al., “Compact underwater single-photon imaging lidar,” Opt. Lett. 50 (6), 1957–1960 (2025). 22. K. Walcarius, J. Rosin, L. Hespel, et al., “Impact of blur on 3d laser imaging: Monte-carlo modelling for underwater applications,” Opt. Express 31(16), 26194–26207 (2023). 23. J. Xian, D. Sun, S. Amoruso, et al., “Parameter optimization of a visibility lidar for sea-fog early warnings,” Opt. Express 28(16), 23829–23845 (2020). 24. T. Factory, “TFC4190-T19,” https://trollfactory.de/Produkte/Silikon-Kautschuk/Haertegrad-Shore/weich-ShoreA25/TFC4191/TFC-Silikon-Kautschuk-Typ-19-transparent-Shore-1-1-Special-Effect-Groesse-1kg-2x500g (2025). 25. AMS-OSRAM, “TFM8828,” https://ams-osram.com/products/sensor-solutions/direct-time-of-flight-sensorsdtof/ams-tmf8828-configurable-8x8-multi-zone-time-of-flight-sensor (2025). 26. A. Pagoli, F. Chapelle, J. A. Corrales-Ramon, et al., “Large-Area and Low-Cost Force/Tactile Capacitive Sensor for Soft Robotic Applications,” Sensors (Basel, Switzerland) 22(11), 4083 (2022). 27. Y. Zhu, Y. Liu, Y. Sun, et al., “Recent advances in resistive sensor technology for tactile perception: A review,” IEEE Sens. J. 22(16), 15635–15649 (2022). 28. Z. Zhang, P. Zhao, P. Lin, et al., “Thermo-optic coefficients of polymers for optical waveguide applications,” Polymer 47(14), 4893–4896 (2006). 29. M. Yazdan Mehr, W. Van Driel, F. De Buyl, et al., “Study on the degradation of optical silicone exposed to harsh environments,” Materials 11(8), 1305 (2018). 30. P. Luesia, M. Crespo, A. Jarabo, et al., “Non-line-of-sight imaging in the presence of scattering media using phasor fields,” Opt. Lett. 47(15), 3796–3799 (2022). Research Article Vol. 33, No. 18 / 8 Sep 2025 / Optics Express 38925 31. D. S. Jeon, A. Meuleman, S.-H. Baek, et al., “Polarimetric iToF: Measuring High-Fidelity Depth Through Scattering Media,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition (IEEE Computer Society, 2023), pp. 12353–12362. 32. J. García-Pueyo, “TimeOfFlightSignalProcessingFTIRTactileSensors,” GitHub (2025), https://github.com/jgarciapueyo/TimeOfFlightSignalProcessingFTIRTactileSensors.