Artificial organic afferent nerves enable closed-loop tactile feedback for intelligent robot

The development of advanced prosthetics, bionic electronics, and intelligent robotic systems draws inspiration from the human tactile system. The human system generates pressure and vibration sensations from spatiotemporal mechanical skin deformations. Afferent nerves carry touch information from mechanoreceptors to the central nervous system (CNS), while efferent nerves carry motor commands from the CNS to muscles. Merkel cells (slow adaptive mechanoreceptors) provide sustained responses to static pressure, while Ruffini corpuscles (fast adaptive mechanoreceptors) perceive directional stimuli and slippage. Neurons communicate through synapses, where chemical fluxes regulate connection strength. Efficient tactile sensing requires distributed and parallel networks of mechanoreceptors, neurons, and synapses. Mimicking the human tactile system has led to the development of capacitive and piezoresistive touch sensors, but these often have volatile behavior, hindering signal processing. Artificial synaptic devices offer analogous functionalities to biological synapses, and while software-based neuromorphic systems exist, they have limitations in emulating the brain's massive parallel processing and power efficiency. Neuromorphic chips using microelectronic devices are emerging, but most lack integration of synaptic activation into a parallel sensory system. Integrating artificial mechanoreceptors with artificial synapses and neurons is key to emulating human tactile sensing and processing. While previous work demonstrated artificial afferent nerves capable of actuation, closed-loop operation for sensing, processing, and actuation is crucial for conscious response and intelligent decision-making in complex tasks. Organic semiconductors are promising for sensing and neuromorphic devices due to their biocompatibility, flexibility, processability, tunability, and low-voltage operation. Organic electrochemical transistors (OECTs), with their large transconductance and low operating bias, are particularly attractive. Their operation, analogous to synaptic excitatory postsynaptic current (EPSC) generation, involves voltage-controlled injection and extraction of mobile ions. However, many OECTs operate in depletion mode, consuming power even without bias, and their use in artificial afferent nerves for tactile information processing is limited. This paper addresses these challenges by proposing a flexible artificial organic afferent nerve (AOAN) using a touch-triggered OECT for hardware-based cognitive recognition in robotics.

Existing research in artificial tactile systems has focused on various approaches to mimic the human sense of touch. Simple two-terminal capacitive and piezoresistive touch sensors have been developed to emulate the signal transduction mechanisms of slow adaptive (SA) mechanoreceptors such as Merkel cells. However, these artificial mechanoreceptors often suffer from volatile behaviors and rapid decay of the temporal state enhancement, which hampers subsequent signal transmission and neuromorphic processing of tactile information. To address this, artificial synaptic devices, mimicking biological synapses, have gained attention. Software-based neuromorphic computing systems have been employed to simulate biological tactile nervous systems, enabling tasks like Braille character discrimination and actuator control in legged robots. However, these software-based approaches are limited by the von Neumann architecture, which struggles to emulate the brain's massively parallel signal processing and power-efficient computation. Hardware-based neuromorphic chips using microelectronic devices offer a solution, but many lack the integration of synaptic activation into a parallel, event-driven artificial sensory system. The integration of artificial mechanoreceptors with artificial synapses and neurons is crucial for building bio-inspired artificial tactile systems. While some research demonstrated the use of artificial afferent nerves to actuate muscles, achieving closed-loop operation for sensing, processing, and actuation is needed for conscious response and intelligent decision-making. Current challenges include the need for low operating voltage, nonlinearity, and biocompatibility in synaptic integration. Organic semiconductors show promise due to their inherent properties, but many existing organic synaptic transistors (OSTs) operate at high voltages and lack long-term state retention. Organic electrochemical synaptic transistors (OESTs), with their coupled ionic/electronic effect, offer a path towards low-power consumption, state retention, and multi-level conductance states ideal for neuromorphic computing, but their application in artificial afferent nerves for tactile information processing remains limited.

This study utilizes an asymmetric regio-regular conjugated polymer, PFT-100, as the channel layer in organic electrochemical synaptic transistors (OESTs). A solid electrolyte (SE) consisting of poly(vinylidene fluoride-co-hexafluoropropylene) (PVDF-co-HFP) and ionic liquid (IL) acts as the ion source for doping the PFT-100 semiconductor. Different ILs with varying anions ([TFSI] and [OTF]) were mixed into the PVDF-co-HFP matrix at different ratios to optimize ionic conductivity and transistor performance. The OESTs were fabricated using a layer-by-layer approach, with a Parylene-C insulating layer separating the source/drain electrodes and the SE. The transfer and output characteristics were measured to determine operating voltage, on/off ratio, threshold voltage (Vth), and transconductance (gm). The memory behavior (on-state current (Ip), rise time (τ), and relaxation time) was evaluated by applying voltage pulses. To mimic synaptic behaviors, voltage pulses were applied at varying frequencies, simulating action potentials. Paired-pulse facilitation (PPF) and post-tetanic potentiation (PTP) were characterized to assess short-term plasticity (STP). Long-term plasticity (LTP) was investigated by repeatedly applying pressure spikes. An artificial organic afferent nerve (AOAN) was constructed by integrating a piezoresistive touch sensor to the gate of the OEST. The pressure-dependent response of the AOAN was characterized, and its sensitivity, relaxation time, and synaptic weight were analyzed. The reversibility of the OEST and AOAN was evaluated by programming them with discrete pulsed voltages or pressures. To demonstrate robotic integration, a dendritic integration function was achieved by integrating four artificial mechanoreceptors in parallel to the OEST. The responses to different spatiotemporal input patterns were analyzed. The AOAN was integrated onto a robot finger to perform tactile information processing and motion recognition. The slip detection and prevention capabilities of the AOAN-integrated robot gripper were tested, using a closed-loop feedback program and a deep learning model (LSTM) to classify various object manipulation activities (gripping securely, slipping, slip prevention). The performance of the AOAN was compared to a system using only a touch sensor without the OEST.

The researchers successfully fabricated organic electrochemical synaptic transistors (OESTs) using PFT-100 as the channel layer and a solid electrolyte (SE) containing ionic liquids. Optimization of the ionic liquid and its blend ratio in the SE resulted in a low threshold voltage (-0.42V) and high transconductance (-1.86mS) for the OESTs. These OESTs exhibited excellent memory behavior with a fast response time (~234ms) and a long relaxation time (~306.5s for 80% TFSI), demonstrating a non-volatile charge retention characteristic significantly longer than previously reported ion-gel-gated synaptic transistors. Integrating the OEST with a piezoresistive touch sensor created an artificial organic afferent nerve (AOAN) that mimics the biological tactile sensory system. The AOAN displayed high sensitivity (106.8 kPa⁻¹) in the low-pressure range (0-30 kPa), crucial for detecting gentle touch. The AOAN showed pressure-dependent and ion-concentration-dependent spiking responses, with higher pressure and higher ion blend ratios resulting in enhanced synaptic weight and longer relaxation times. Short-term plasticity (STP) was emulated using the AOAN, demonstrating paired-pulse facilitation (PPF) and post-tetanic potentiation (PTP) behaviors similar to biological synapses. Long-term plasticity (LTP) was also observed in the AOAN, with the memory level increasing with the number of pressure spikes and applied pressure. A training process using repeated pressure spikes enhanced the AOAN's sensitivity to subtle touch, allowing it to detect even weak stimuli that were undetectable by the touch sensor alone. Dendritic integration was demonstrated by integrating multiple touch sensors to the OEST, enabling the AOAN to perceive directional tactile stimuli and recognize object slippage. Integration of the AOAN into a robotic gripper with a closed-loop feedback program enabled real-time slip detection and prevention. The robot successfully recognized and prevented slippage during various object manipulations (gripping, slippage, regripping, shaking), demonstrating the practical application of the AOAN in intelligent robotics. Deep learning using an LSTM model showed significantly improved accuracy (98.7 ± 0.3% after 10 epochs, 99.0 ± 0.3% after 50 epochs) in classifying different object manipulation activities when using the AOAN compared to using only the touch sensor (75.2 ± 0.5% after 10 epochs, 97.1 ± 0.3% after 50 epochs).

This study successfully demonstrated the feasibility and advantages of using an artificial organic afferent nerve (AOAN) based on organic electrochemical synaptic transistors (OESTs) for creating intelligent robots with advanced tactile feedback. The AOAN's ability to mimic key aspects of biological tactile sensing, including short-term and long-term plasticity, sensitivity to subtle touch, and dendritic integration, represents a significant step forward in neuromorphic computing. The experimental results, particularly the high accuracy of slip detection and prevention in the robotic gripper, showcase the practical potential of the AOAN for various applications requiring advanced tactile interaction and feedback. The low operating voltage (-0.6V) and excellent stability of the OESTs are crucial for creating energy-efficient and reliable devices. The integration of multiple touch sensors with the OEST enables efficient processing of complex tactile information. The use of deep learning to process the spike-encoded signals further enhances the system's ability to differentiate complex tactile patterns and to learn and adapt to different manipulation scenarios. This work significantly contributes to the field of biomimetic electronics and opens new avenues for developing next-generation intelligent neurorobotics.

This paper presents a novel artificial organic afferent nerve (AOAN) based on a pressure-activated organic electrochemical synaptic transistor. The AOAN demonstrates low operating voltage, high sensitivity, and the ability to mimic short-term and long-term plasticity, enabling advanced tactile feedback for intelligent robots. The successful integration of the AOAN into a robotic gripper with closed-loop control demonstrates real-time slip recognition and prevention. This research significantly advances the development of biomimetic electronics and intelligent neurorobotics, paving the way for more sophisticated and adaptable robotic systems. Future research could explore expanding the array size for improved spatial resolution, investigating different materials for enhanced performance, and integrating the AOAN into more complex robotic systems for a wider range of applications.

While this study demonstrates significant progress in developing an artificial organic afferent nerve (AOAN) for robotics, several limitations should be considered. The current implementation of the AOAN uses a relatively small array of artificial mechanoreceptors, limiting its spatial resolution. This could affect the accuracy of tactile pattern recognition, particularly in scenarios involving complex textures or fine manipulation tasks. Further research is needed to expand the array size and improve spatial resolution. The study mainly focuses on the detection of slip motion. More complex tactile stimuli, such as different textures or vibrations, should be investigated. Finally, the long-term reliability and durability of the AOAN under continuous operation need further evaluation for practical applications.