Neuromorphic tactile intelligence in robotics and prosthetics

Recreating the human sense of touch in artificial systems remains a key challenge. We developed flexible multilayer tactile sensors, a biomimetic prosthetic hand, feedback interfaces, and neuromorphic algorithms to emulate biological sensing and perception. By merging soft materials with computational neural models, we are advancing perception in robotics and improving haptic feedback in prosthetic systems.

Focus Group: Human-Centered Neuroengineering

Prof. Nitish V. Thakor (Johns Hopkins University), Alumnus Hans Fischer Senior Fellow (funded by the Siemens AG) Fengyi Wang (TUM), Doctoral Candidate
Host: Prof. Gordon Cheng (TUM)
image: CC BY

Figure 1

Fig. 1, Hybrid robotic hand integrating soft mechanics and neuromorphic sensing. The hybrid hand combines a rigid skeletal structure for precision and force control with soft joints for compliance and safety. Multilayer tactile sensors emulate human mechanoreceptors, converting pressure and vibration into spike-based neuromorphic signals. This integration enables dynamic manipulation, touch perception, and reflexive control inspired by human sensory processing.
source: CC BY

Research concept and goals:

Despite rapid progress in sensing and actuation, robotic manipulation still falls short of the human hand in sensitivity and adaptability. Conventional tactile sensors provide static measurements of force or temperature but lack the speed and accuracy of biological touch. Neuromorphic computing provides an alternative by using event-driven spike communication similar to neuronal signaling, allowing faster perception with lower power requirements. In upper-limb prosthetics, the absence of reliable somatosensory feedback limits the function and embodiment. Prior efforts have improved task performance, but they often remain constrained by bulky or tethered hardware, limited channel count, and inconsistent real-time operation [1].

We integrate these separate efforts within one neuroengineering framework that connects soft-robotic design, neural modeling, haptic feedback, and neuromorphic algorithms. It aims to translate the principles of the human nervous system into robotics and prosthetics through neuromorphic circuits and spiking neural networks that process information in a brain-like manner. The goal is to create artificial limbs and humanoid robots that can sense, interpret, and react to their environment with human-like perception. The overarching vision is to close the sensory loop between humans and prosthetics through technologies and create intelligent robotic systems that behave and learn as humans do.

Findings

During the Fellowship, the team integrated soft-robotic design, computational modeling, and robotic implementation to develop a biologically grounded framework that links tactile sensing, reflexive control, and higher-level perception in robots, while improving haptic feedback for human users. Flexible multilayer tactile sensors were developed to replicate the distribution of mechanoreceptors in the human fingertip. Each layer responded to specific stimuli, from static pressure to high-frequency vibration, and when embedded in a soft robotic finger, the sensors enabled accurate identification of materials and stiffness during manipulation.

These sensors were integrated into a biomimetic prosthetic hand to achieve precise and adaptive grasping. The hand incorporates a rigid internal skeleton for structural strength and soft joints that ensure compliant and safe interaction. This integration enables the hand to perceive contact forces and texture in real time, allowing stable grasping of delicate or irregular objects.

A neuromorphic palpation algorithm was implemented to extend tactile perception to robotic exploration. Using biologically inspired encoding, a robotic finger accurately identified bone fractures while remaining invariant to palpation speed, reproducing a fundamental feature of human tactile sensing and suggesting new possibilities for autonomous robotic inspection.

The Focus Group also demonstrated that biologically inspired nociception can be implemented in a humanoid robot to provide autonomous self-protection. The system equips the robot with an artificial nociceptive reflex arc that generates graded withdrawal movements in response to noxious heat. The reflex reproduces key characteristics observed in human physiology, confirming its biological plausibility.

We further advanced multimodal sensing and perception by integrating multiple sensors of different modalities into a soft anthropomorphic hand to enable real-time object and pose recognition. The system fuses force, vibration, and deformation cues with neural encoding to generate spike-based representations of contact events. These spatiotemporal patterns allow the hand to classify object identity and estimate pose during manipulation with high accuracy.

In parallel, a closed-loop sensory feedback system for prosthetic applications was developed. Tactile data were converted into neuromorphic spike trains and transmitted through vibrotactile or transcutaneous electrical stimulation to the user’s residual limb. The stimulation intensity and frequency changed dynamically according to contact force and texture, producing a natural, continuous sensation of touch.

Collectively, these results show how biological understanding can lead to practical, human-centered engineering solutions in robotics and prosthetics.

Collaboration and knowledge exchange

Beyond our laboratory research, we promoted international collaboration through two scientific events. The first, a two-day workshop on neuromorphic computing held in October 2023, provided a forum for researchers, doctoral candidates, and students to discuss advances in neuromorphic systems, hardware, and algorithms. The second workshop, on sensory integration in neuroprosthetics and rehabilitation, took place in October 2024, bringing together experts in neuroscience, robotics, and biomedical engineering to explore how sensory inputs can be merged and used coherently in humans, prosthetic systems, and robots. Both events fostered dialogue across disciplines and highlighted recent progress as well as open challenges in achieving human-like sensory integration and perception.

Figure 2

Fig. 2, Bio-plausible nociceptive reflex on a humanoid robot. A spiking neural circuit controlling the robot’s arm reproduces the heat-evoked nociceptive withdrawal reflex. Temperature sensors detect noxious heat and generate neuromorphic spike signals processed by interneuron and motor-neuron models, producing graded withdrawal movements in real-time.
source: CC BY

Figure 3

Fig. 3, Group photo taken during the 2024 Workshop on Sensory Integration in Neuroprosthetics and Rehabilitation, held at TUM
source: CC BY

Fig. 4, Prosthetic system with tactile feedback. The platform integrates a sensory glove, a bypass prosthetic interface, and a vibrotactile armband. Contact forces detected by the glove are wirelessly transmitted to the armband, where vibrotactile actuators reproduce touch sensations on the user’s skin. This closed-loop setup allows simultaneous grasp control and tactile feedback, supporting more intuitive prosthetic operation.
source: CC BY

Outlook:

The systems we developed will contribute to the next generation of prosthetic limbs, providing continuous, natural haptic feedback. Neuromorphic palpation brings human-like tactile exploration to robots, enabling precise sensing of contact dynamics and material properties for advanced manipulation and control. More broadly, the project demonstrates how neuroscience and engineering can be integrated into a unified framework for human-centered robotics, laying the foundation for lifelike robotic systems that share not only human motion but also human-like perception and intelligence.

Acknowledgment:

We acknowledge support from the BMBF-NSF-funded COMMI Project (Computational Models of Multisensory Integration of the Upper Limbs in Humanoids and Amputees), which provided essential resources and collaboration for the neuromorphic modeling and sensory-integration studies reported here.

[1]
Y. Angkanapiwat, A. Slepyan, R. J. Greene and N. Thakor, “SensoPatch: A Reconfigurable Haptic Feedback with High-Density Tactile Sensing Glove,” 2024 IEEE Biomedical Circuits and Systems Conference (BioCAS), Xi’an, China, (2024).

Selected publications

A. S. Pimpalkar, A. Slepyan, and N. V. Thakor, “Vibrations at First Contact Encode Object Stiffness Before Grasp Completion,” IEEE Sens. Lett., vol. 9, no. 8, pp. 1–4, Aug. 2025.
S. Bello, M. M. Iskarous, S. Sankar, and N. V. Thakor, “Robotic Palpation of Fractures Using Bioinspired Tactile Sensor and Neuromorphic Encoding Algorithm,” IEEE Trans. Med. Robot. Bionics, vol. 7, no. 3, pp. 1175–1185, Aug. 2025.
S. Sankar et al., “A natural biomimetic prosthetic hand with neuromorphic tactile sensing for precise and compliant grasping,” Sci. Adv., vol. 11, no. 10, pp. eadr9300, Mar. 2025.
M. M. Iskarous et al., “Invariant neuromorphic representations of tactile stimuli improve robustness of a real-time texture classification system,” Advanced Intelligent Systems, vol. 7, no. 10, pp. 2401078, Apr. 2025.
F. Wang, J. R. G. Olvera, N. Thakor, and G. Cheng, “A Bio-Plausible Approach to Realizing Heat-Evoked Nociceptive Withdrawal Reflex on the Upper Limb of a Humanoid Robot,” IEEE Robot. Autom. Lett., vol. 8, no. 6, pp. 3398–3405, June 2023.

Full list of publications