Sep 2026· Coins· pp. 1-6· 0 citations· 24 references
Computer Science
TL;DR
The feasibility of deploying multi-task tactile perception on modern NPU-enabled embedded platforms for collaborative robotic applications is demonstrated, corresponding to an approximately 62× speedup over a benchmark STM32F7-based implementation.
Abstract
Edge Artificial Intelligence (Edge AI) enables intelligent embedded systems to execute deep neural networks locally, reducing latency and eliminating the dependence on cloud computing. In this context, collaborative robotics represents a particularly challenging application domain, where tactile perception has to satisfy strict real-time constraints on resource-constrained hardware. This work presents an Edge AI system for real-time tactile perception based on an artificial skin equipped with 21 Fiber Bragg Grating (FBG) sensors. A compact multi-task neural network jointly estimates contact location, applied force, and contact state from multiplexed optical measurements. The network is optimized through quantization-aware training and deployed as an INT8 TensorFlow Lite model on an STM32N6 microcontroller, exploiting the integrated Neural-ART neural processing unit for hardware-accelerated inference. Experimental evaluation demonstrates a force RMSE of 0.57 N, a median localization error of 3.8 mm, a contact-state classification accuracy of 97.9%, and an inference time of 0.35 ms, corresponding to an approximately 62× speedup over a benchmark STM32F7-based implementation. These results demonstrate the feasibility of deploying multi-task tactile perception on modern NPU-enabled embedded platforms for collaborative robotic applications.
This work establishes a new material-architecture paradigm that unifies ultrarobustness, dual-modality, and in-sensor computing, setting the stage for next-generation wearable human-machine interfaces with both comfort and intelligence.
Hua-Sen Wang, Lei Liu, Shi-Fan Yu et al.· Advances in Materials· 0 citations
Tactile sensing is increasingly being incorporated into learning-based robotic manipulation, yet many existing approaches rely on spatially distributed sensors. Here we introduce {SpectRobot}, a framework that transforms single-point tactile signals into compact time-frequency spectrograms. These spectrograms encode hi...
Joseph Rigal, E. Virot, Caroline Pascal· arXiv.org· 0 citations
Tactile-JEPA is an efficient self-supervised pre-training method that uses the spatial arrangement of tactile sensors to learn topology-aware representations, and is trained to predict the embeddings of masked sensing elements from the unmasked remainder, using the sensor connectivity graph to guide spatial masking.
E. Kovtun, M. Konovalov, Andrey Sakhovskiy et al.· arXiv.org· 1 citation
This paper presents a distributed Edge AI architecture for the real-time, contactless teleoperation of a 6-DOF robotic manipulator using natural hand gestures. In environments with strict sterility or safety requirements, such as cleanrooms or hazardous industrial zones, traditional physical interfaces are impractical....
B. Amirkhanov, Miras Tokhtassyn, G. Amirkhanova et al.· Asia Conference on Edge Inte...· 0 citations
Real-time object perception with geometric awareness is a critical enabling technology for intelligent visual perception in robotics. However, deploying robust visual recognition on resource-constrained edge devices remains challenging due to computational limitations and the lack of geometric priors. This paper presen...
Zong-Wen Li, Yu-Kun Wei, Zhi-Gang Zhou et al.· International Conference on...· 0 citations
This article showcases the architectural bottlenecks, material science challenges, and embedded computing requirements that influence whether E-skin can transition from laboratory prototypes to deployable robotic systems, and claims that the next E-skin trends will be built on top of heterogeneous architectures compose...
Ioannis Papadongonas, Alexandros Gazis, Vasileios Fanidis et al.· Academia Engineering· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.