Bioinspired adaptive response speed for high-quality human-robot interactions

Abstract

In human-robot interaction (HRI), the quality of user experience is paramount. Thus, developing strategies to tailor robotic responsiveness to user comfort zone is essential. However, current methods remain constrained by complex software and limited to unidirectional speed enhancement. Inspired by the adaptive pathways of biological brain, this study introduces a photochromic diarylethene–doped organic floating-gate field-effect transistor (OFGFET) capable of bidirectional response speed control. Mechanism investigations revealed that a light-controllable back charge transfer process facilitates dynamic speed adjustments. Notably, the OFGFET device potentially enables effective regulation of response speeds to enhance user comfort while maintaining efficient and detailed movie playback. This work establishes a distinctive paradigm in HRI, offering unprecedented adaptability that could transform user experience design across diverse interactive technologies.

A bioinspired device adaptively controls the response speeds via light and achieves high-quality human-robot interactions.

INTRODUCTION

Human-robot interaction (HRI) technology has gained much attention with the rapid advancements in artificial intelligence, particularly in humanoid robots (1–4). HRI explores the dynamic relationship between systems and users to achieve high interaction quality, ensuring that robots effectively understand and respond to human needs and preferences (2, 5). Machine feedback is critical for enhancing user experience, addressing subjective human perceptions in areas such as response speed (6, 7), video output quality (8–10), and other key factors (11–15). Specifically, a robot’s response speed must remain within a consistent range; excessively rapid responses can induce discomfort, while overly delayed reactions may appear disengaging (16, 17). Consequently, the adjustability of robotic response speed has become a priority. Numerous approaches have been explored to improve robotic response speeds (4, 18, 19), such as using cerebellar models to regulate nondeterministic time delays in control systems or using sensory manipulation to mitigate subjective perceptions of delay (7, 20). However, these strategies typically rely on complex software integration and are limited to unidirectional speed control, lacking the flexibility for bidirectional adjustments.

In the human brain, two switchable pathways—cortical processing streams and subcortical routes—mediate responses to external stimuli (21–24). Typically, external information is processed through cortical pathways, engaging higher-order brain regions responsible for complex cognitive functions, such as perception, decision-making, and planning. In contrast, during imminent threats, the brain rapidly shifts to subcortical routes, modulating fast-acting neuropeptides and neurotransmitters to enable immediate responses (Fig. 1A) (25–28). This dual-pathway system, governed by hypothalamic-pituitary-adrenal axis control, ensures both efficiency and robustness in adapting to dynamic environments.

Fig. 1. Two regulatory pathways in the human brain and OFGFET device.

(A) Cortical visual stream (white) and subcortical processing stream (red) of the human brain under two different reactions versus the device with adjustable response speed. LGN, lateral geniculate nucleus; V1, primary visual cortex; ES, extrastriate cortex; IT, inferotemporal cortex; vmPFC, ventromedial prefrontal cortex; PAG, periaqueductal gray; SC, superior colliculus. (B) Schematic diagram of OFGFET devices and structures of molecules. Vis, visible.

Inspired by the responsive mechanism of the human brain, we developed an organic floating-gate field-effect transistor (OFGFET) to bidirectionally control response speeds. Upon constructing a floating gate by doping photoisomeric compounds into polymethyl methacrylate (PMMA), we can modulate the back charge transfer rate by altering the trapping ability of the floating gate upon varying light [ultraviolet (UV) or visible] irradiation. Systematical characterization revealed that UV light decreased the back charge transfer rate by lowering lowest unoccupied molecular orbital (LUMO) energy levels of photoisomeric compounds, extending the response time, while visible light increased the back charge transfer speed upon increasing LUMO energy levels, shortening the response time. As a result, these devices potentially enable the adaptive control of robot response speeds in HRI.

RESULTS

Design and fabrication of the device

An OFGFET was fabricated with the structure shown in Fig. 1B. A newly synthesized photoisomeric DAE1 compound was dispersed in a PMMA film, serving as a floating gate to trap charges. The polymer PDPP2TBT (29), known for its high hole transport mobility, was spin coated onto the PMMA:DAE1 film to form the conductive channel for charge carrier transport (30, 31). The ring-open isomer of DAE1 (DAE1_o) can transition to its ring-closed form (DAE1_c) under UV light, which reverts to DAE1_o under visible light (fig. S1) (32–36). The PDPP2TBT layer exhibited a transmittance exceeding 60% in the 300- to 600-nm range (fig. S2), demonstrating that the photochromic reaction of DAE1 can be effectively induced by photoirradiation even with the PDPP2TBT layer overlaying it. In addition, we use orthogonal solvents in the preparation process to ensure that the spin-coating process does not damage the layer underneath (fig. S3).

The transfer curves of DAE1_o- and DAE1_c-based devices demonstrated that the source-drain current (I_DS) increased with increasing negative gate voltage (V_G), exhibiting typical p-type characteristics (Fig. 2A). In contrast, I_DS showed a minimal change under positive V_G (fig. S4), indicating that the floating gate predominantly exhibited electron trapping rather than hole trapping capabilities (31, 37, 38).

Fig. 2. Photocontrolled response performance of the devices.

(A) Transfer characteristics of DAE1-based devices in two states. (B) Normalized I_DS curves of DAE1_o- and DAE1_c-based devices. a.u., arbitrary units. (C) Definition of response index τ_p. (D) Δτ_p of devices with different ratios of PMMA:DAE1. (E) Normalized I_DS decay curves under different V_pulse durations. (F) τ_p varies with pulse width in DAE1_c- and DAE1_o-based devices. (G) τ_p values under different electrical pulse numbers. The error bars indicate the standard deviation of three tests. (H and I) Normalized I_DS decay curves under different V_G values in (H) DAE1_c- and (I) DAE1_o-based devices.

Furthermore, under the same V_G, the I_DS of DAE1_o-based devices was higher than that of the DAE1_c-based one, suggesting that more electrons were trapped in the PMMA:DAE1_o layer, thereby enhancing gate tunneling (39). In comparison, the transfer curves of transistors without DAE1 doping showed no changes following UV or visible light irradiation (fig. S5), confirming that the conductivity changes in DAE1-doped devices arose from the photoisomeric reaction of DAE1 molecules.

Light-controllable response of the device

To investigate the response characteristics of the OFGFET, five successive voltage pulses (V_pulse = 10 V, 1 s) were applied to the gate electrode. In both DAE1_o- and DAE1_c-based devices, I_DS increased progressively with each pulse (Fig. 2B and fig. S6), indicating efficient charge trapping. After five pulses, I_DS decayed in a single-exponential form before stabilizing at a prolonged retention phase. Notably, DAE1_c-based devices exhibited a slower decay and longer retention time compared to their DAE1_o-based counterparts.

To quantify the decay speed following the removal of electrical pulses, the I_DS curves measured 200 s postpulse were normalized (Fig. 2C) and fitted using a single-exponential function. The response index (τ_p) was defined as follows

$$I=I_0\text{exp}(-t/{\mathrm{\tau }}_{\mathrm{p}})$$ (1)

where I₀ represents the initial current value, and τ_p denotes the time required for I_DS to decline to 1/e of its initial value. A smaller τ_p corresponds to a faster response speed. For instance, the τ_p values of DAE1_o- and DAE1_c-based devices were 30.78 and 45.97, respectively, indicating a faster response speed for the DAE1_o-based device. These findings suggest that the response speed can be readily bidirectionally tuned through UV or visible light irradiation.

The differences in τ_p (Δτ_p) between DAE1_o- and DAE1_c-based devices reflect varying levels of regulatory capability, with a larger Δτ_p indicating a broader adjustable range. To optimize conditions for a large Δτ_p, the mass ratio of DAE1:PMMA was systematically varied. Δτ_p initially increased with the DAE1:PMMA ratio, peaking at 10.83 when the ratio reached 4:6 (Fig. 2D and fig. S7). Beyond this ratio, Δτ_p began to decline, decreasing to 9.18 at a 5:5 ratio. Therefore, a 4:6 ratio was selected as the optimal condition for subsequent experiments.

Upon increasing the electrical pulse width from 1 to 9 s, the Δτ_p of both DAE1_c- and DAE1_o-based devices increased (Fig. 2E), indicating that a longer pulse duration enhances the rate difference between the switching loops. As shown in Fig. 2F, τ_p values for both the closed- and open-ring states exhibited exponential decreases with increasing pulse duration, suggesting that longer electrical stimuli trap more charges in the DAE1:PMMA layer, resulting in a smaller τ_p after the removal of the V_pulse. Similarly, increasing the number of pulses induced more trapped charges, leading to a reduced τ_p (Fig. 2G).

In addition, τ_p demonstrated distinct voltage-dependent characteristics, decreasing as the V_pulse (Fig. 2, H and I) or V_G (fig. S8) increased, as more charges were induced and trapped. To finely modulate τ_p, different V_pulse and V_G values were applied. As illustrated in Fig. 3A and fig. S9, the upper and lower semicircles represent the performance of DAE1_o- and DAE1_c-based devices, respectively. The segmented sectors correspond to different V_pulse values, which increased sequentially from 5 to 10 V in a clockwise direction. The concentric rings represent different V_G values, ranging from −5 to −10 V from the inside to outside. Notably, the τ_p of the DAE1_o device was smaller than that of the DAE1_c-based device under the same voltage conditions, indicating a stronger charge trapping ability of the floating gate. Moreover, τ_p gradually decreased with V_pulse or V_G increasing, resulting in a faster response speed. To assess the repeatability and stability of these devices, τ_p was measured over 10 repeated UV and visible light irradiation cycles (Fig. 3B). The results showed minimal variation, suggesting excellent stability. Furthermore, OFGFET devices were fabricated on flexible polyethylene terephthalate (PET) substrates (fig. S10). The I_DS exhibited similar behaviors to those on rigid substrates under electrical pulses (Fig. 3C). As shown in fig. S11, photographs of the device under various bending degrees demonstrated that the device retained its structural integrity, with no visible signs of delamination or detachment. In addition, τ_p barely changed under different bending angles (fig. S12), which indicated excellent mechanical robustness and strong potential for integration into flexible electronic systems.

Fig. 3. Electrical performance of the device.

(A) Modulated τ_p under different V_G and V_pulse values. (B) τ_p under 10 UV/visible light cycles. (C) Normalized I_DS curves of DAE1_o- and DAE1_c-based devices on the PET substrate.

Mechanism of photocontrolled response speed

To investigate the mechanism underlying photocontrolled response speed, a series of characterizations was performed. First, the HOMO (highest occupied molecular orbital) and LUMO energy levels of DAE1_c, DAE1_o, and PDPP2TBT were calculated to be −5.51/−3.55, −5.49/−2.00, and −5.42/−4.07 eV, respectively, on the basis of cyclic voltammetry measurements (fig. S13), ultraviolet photoelectron spectroscopy (fig. S14), and optical absorption spectroscopy. Notably, the LUMO of PDPP2TBT is lower than those of both DAE1_c and DAE1_o, meaning that electron transfer from PDPP2TBT to DAE1 requires an applied voltage, whereas the reverse process occurs spontaneously (38). As demonstrated in fig. S15, the surface morphology remained statistically invariant under both illumination modes, with root-mean-square roughness (R_q) values of 4.15 nm (DAE1_o) and 4.08 nm (DAE1_c). These findings exclude surface morphology as a variable in the observed light-gated trap state engineering. Electrostatic force microscopy was used to investigate the changes in surface charge distribution of PMMA:DAE1 thin films following exposure to visible and UV light. As shown in fig. S16, no notable changes in the surface charge distribution were observed upon illumination. Therefore, it can be concluded that the interface trap states are not affected by light exposure.

To investigate the differences in photocurrent attenuation between DAE1_c and DAE1_o, we first examined the charge injection barriers. A device with a structure of ITO (indium tin oxide)/PDPP2TBT/PMMA:DAE1/Au was fabricated to measure capacitance-voltage (C-V) curves (fig. S17). As shown in Fig. 4A, when the voltage shifted from negative to positive, the capacitance of both devices initially increased and then decreased. The initial increase in capacitance indicates carrier injection into the devices, while the subsequent decrease is attributed to enhanced radiative recombination as the applied voltage increases. Notably, the DAE1_c-based device exhibited a faster capacitance rise rate than the DAE1_o-based device, suggesting enhanced carrier injection in the DAE1_c-based device (40). Furthermore, the peak capacitance value (C_p) of the DAE1_c-based device (≈0.14 nF) was larger than that of the DAE1_o-based device (≈0.12 nF). In addition, the voltage corresponding to the C_p of the DAE1_c-based device (≈0.32 V) was higher than that for the DAE1_o-based device (≈0.17 V). These observations indicate that the charge injection barrier between PDPP2TBT and DAE1_c is higher than that between PDPP2TBT and DAE1_o (41, 42).

Fig. 4. Mechanism of photocontrolled response speed.

(A) C-V curves of the device after UV and visible light irradiation. During the measurements, the modulating frequency is 1 kHz. (B) Normalized V_cpd under electric pulses. Inset: exponential fitted V_cpd decay. (C and D) Energy levels under different conditions. (i) V_G < 0 V. (ii) V_G > 0 V. The electrons tunneled to the DAE1 floating gate. (iii) V_G = 0 V. The electrons stored in DAE1 slowly released to PDPP2TBT. (E) τ_p, Δτ_p, and ΔE of DAE1, DAE2, and DAE3. The error bars indicate the standard deviation of three independent devices.

To further investigate the mechanism, in situ Kelvin probe force microscopy was used to monitor real-time changes in surface potential (V_cpd) during the process (fig. S18A). When a positive pulse voltage was applied, V_cpd increased, indicating electron tunneling into the PMMA:DAE1 layer (fig. S18B). Upon the removal of the pulse voltage, V_cpd slowly decayed, suggesting that trapped electrons were gradually released to PDPP2TBT (43, 44). By fitting the decay curve with a single exponential function, the lifetime of V_cpd for the DAE1_c-based device was calculated to be 451.98 s, which was longer than that of the DAE1_o-based device (382.98 s) (Fig. 4B and fig. S18C), consistent with the observed order of τ_p.

Accordingly, we propose the working mechanism as shown in Fig. 4 (C and D). When a positive V_pulse is applied, electrons in PDPP2TBT tunnel into the DAE1 floating gate layer. After the V_pulse is removed, the stored electrons in the floating gate slowly back-transfer to the PDPP2TBT layer under the driving force of the LUMO energy level difference (ΔE_LUMO) between DAE1 and PDPP2TBT. The ΔE_LUMO determines the rate of back charge transfer according to the Fermi golden rule (45, 46), which ultimately influences the τ_p of the device (47, 48). Thus, a larger ΔE_LUMO between DAE1_o and PDPP2TBT leads to a faster back charge transfer rate and a smaller τ_p.

To investigate the generality of the back charge transfer mechanism, two DAE1 derivatives with different side chains (DAE2 and DAE3 shown in figs. S19 and S20) were synthesized. On the basis of cyclic voltammetry measurements (fig. S21), the LUMO energy level differences (ΔE) between DAE_c and DAE_o of DAE2 and DAE3 were calculated to be 1.00 and 1.50 eV, respectively, both smaller than that of DAE1 (1.54 eV). Not surprisingly, both DAE2 and DAE3 based OFGFET devices exhibited p-type characteristics (fig. S22). Upon application of the pulse voltage, the DAE_o-based devices exhibited lower τ_p than the DAE_c-based devices, suggesting the faster back charge transfer of DAE_o-based devices (Fig. 4E). The Δτ_p of the DAE2-, DAE3-, and DAE1-based devices increased from 8.02 and 10.27 to 15.19, consistent with the order of ΔE values. Given that the E_LUMO of PDPP2TBT was fixed, the ΔE_LUMO of diarylethene (DAE) determined Δτ_p. These observations suggest that the response range can be precisely tuned through molecular engineering via the back charge transfer mechanism.

Photocontrolled response speed in HRI

In the process of HRI, it is crucial for robots to maintain a comfortable response speed. A response that is too fast may make humans feel unwelcome, while an overly slow response can give the impression of delay (16, 17). As shown in Fig. 5A, the τ_p value for the human comfort zone is set between 35.5 and 53.5, while those τ_p values out of this range are considered to be uncomfortable. When V_G and V_pulse inputs are fixed, the τ_p values of both DAE1_o- and DAE1_c-based devices can be tuned via UV or visible light irradiation. This enables the adjustment of the response speed from the uncomfortable region into the comfortable zone. For example, when V_G was −9 V, the τ_p of DAE1_o-based devices was 33.97, which fell outside the comfortable zone for humans. To tune the τ_p within the comfortable zone, UV light irradiation was applied, resulting in a τ_p of 39.43, given that DAE1_o was converted to DAE1_c. Similarly, when V_G was −6 V, the τ_p of the DAE1_c-based devices was 53.64, which was too slow for humans. Upon visible light irradiation, the τ_p decreased to 44.00, within the comfortable zone, as DAE1_c was converted to DAE1_o.

Fig. 5. Application in HRI.

(A) Various response speeds of DAE1_o- and DAE1_c-based devices under different V_G values (V_pulse = 10 V). Different ranges of τ_p correspond to different perceptual intervals. The error bars indicate the standard deviation of three tests. (B) Adjustment of response speeds into the comfortable speed zone under a series of (V_G, V_pulse) actions. (C and D) Accelerated playback speed of a snail crawling movie with the device after visible light irradiation. (C) Original 16 frames within 4 s and the final frame and (D) accelerated processed 50 frames within 4 s and the final frame. (E and F) Decelerated playback speed of a mosquito flight film with the device after UV irradiation. (E) Incomplete flight trajectory of mosquitoes in the original movie and (F) detailed flight trajectory of a fly in the decelerated movie.

Robots are often required to perform multiple tasks simultaneously, and the response times of these tasks are varied (49, 50). To simulate these processes, various V_G and V_pulse conditions were applied to the device, where each (V_G, V_pulse) condition represented a distinct command for the robot to execute. Under each (V_G, V_pulse) condition, the τ_p value may fall within or outside the comfortable region (Fig. 5B). For example, when the conditions (−9, 7), (−10, 8), (−10, 10), (−7, 7), (−7, 8), and (−10, 9) were input, the τ_p values of these tasks were 59.11, 38.31, 30.08, 63.40, 55.20, and 32.32, respectively. Obviously, most of these values fell outside the comfortable range. However, after suitable light irradiation, τ_p values were tuned to be 43.56, 55.97, 35.77, 53.43, 52.15, and 48.13, respectively, which are within the comfortable zone.

During HRI, adjusting video playback speed is crucial for an intuitive and efficient experience (51, 52). To demonstrate the capability to modulate speed, the device controls the video frame output via electrical pulses. Each frame is displayed when the current decays to 98% of its peak value, with different decay rates corresponding to varying playback speeds. For instance, in a video showing a snail crawling (movie S1), only 16 frames were played in 4 s at the original speed, resulting in the snail covering only a short distance (Fig. 5C). This slow speed made the video feel redundant and unengaging. To enhance the user’s experience, visible light was applied to activate the “fast channel,” enabling 50 frames to be played in the same 4-s interval. This made the snail appear to crawl much farther (Fig. 5D and movie S2), substantially enhancing the efficiency of information transmission.

For fast-moving subjects, such as mosquitoes in flight, their motion often exceeds the resolution limits of the human eye, making it difficult to track their flight paths (53, 54). Thus, slowing down the video playback can help reveal these details in HRI. When the DAE1_o-based device was used to observe mosquito flight traces (movie S3), only partial trajectories were visible initially (Fig. 5E). However, when UV light switched the device to the “slow channel,” the detailed flight path became clearly visible, allowing for easier observation (Fig. 5F and movie S4).

DISCUSSION

In conclusion, drawing inspiration from the brain’s adjustable response mechanisms, we developed an OFGFET device with light-controllable response speed, enabled by the photoisomeric floating gate. Upon UV or visible light irradiation, the response rate of OFGFET can be bidirectionally tuned to achieve a comfortable zone for HRI. Mechanistic investigations revealed that the ΔE_LUMO between DEAs and DPP modulates injection barriers and back charge transfer rates, enabling the precise control of response dynamics. Notably, the bidirectional tunability can be readily adjusted through molecular engineering, facilitating dual control over robotic response and video playback speeds. This work presents an innovative strategy for achieving the bidirectional control of response speed, offering an efficient pathway to achieve the high adaptability and functionality of intelligent robots in diverse HRI scenarios.

MATERIALS AND METHODS

Materials

The semiconductor polymer PDPP2TBT was purchased from Solarmer [M_w (weight-average molecular weight) > 15,000]. The DAEs were synthesized by the method in figs. S23 to S25 and characterized (figs. S26 to S28).

Fabrication of the devices

The Si substrate with 200-nm SiO₂ was cleaned with deionized water, alcohol, and acetone under ultrasound for 30 min, followed by drying with nitrogen blow. The DAEs and PMMA were dissolved in chloroform (5 mg ml⁻¹) in varying proportions and stirred for ~10 hours. After oxygen plasma treatment, the mixed solution of DAE1 and PMMA was spin coated (3500 rpm, 45 s) on the Si/SiO₂ wafer. Then, a PDPP2TBT solution (10 mg/ml in toluene) was then spin coated on top of the film at 2000 rpm for 60 s and annealed at 120°C for 10 min. Subsequently, the Au electrode (60 nm) was deposited by thermal evaporation (pressure of <10⁻⁴ Pa; evaporation rate, 0.05 nm s⁻¹) through the shadow mask.

Fabrication of the flexible devices

The PET substrate (0.125 mm in thickness) with an ITO electrode (15 ohms, 135 nm) was cleaned as described above. PMMA was dissolved in acetone (20 mg ml⁻¹), stirred for 10 hours, and then spin coated onto the PET substrate at 2500 rpm for 45 s (twice). The floating-gate layer and channel were fabricated following the same procedure as described above.

Electrical characterization

All electrical properties were performed using a Keysight B1500A in ambient air. The pulse voltage was generated by the UTG4122A function/arbitrary waveform generator and applied to the gate electrode via a probe.

Surface morphology and charge measurement

Atomic force microscopy morphology, electrostatic force microscopy, and in situ Kelvin probe force microscopy measurements were conducted in air using a Bruker Fastscan AFM instrument. The device was fixed on a glass sheet, and three electrodes were led out through gold wire. The pulse voltage was applied to the gate through an Agilent B2920, while the source and drain electrodes were grounded. The probe measured the variation of surface potential at a distance of 100 nm from the surface.

Video playing speed adjustment by OFGFET

In the video playback system, each frame is controlled by an electrical pulse. For example, in a video with 50 frames, this corresponds to 50 pulsed inputs. The current generated by each pulse is monitored: When the current reaches its maximum value, it is normalized to 1. A frame is displayed when the current decays to 0.98, which is considered the threshold for triggering the output of one video frame.

Supplementary Materials

The PDF file includes:

Supplementary Text

Figs. S1 to S28

Legends for movies S1 to S4

Other Supplementary Material for this manuscript includes the following:

Movies S1 to S4