Neuromorphic compression and telemetry ASICs for humanoid tactile sensing applications

The sense of touch is essential for humans to safely and dexterously interact with their environment, yet remains one of the least mature sensing modalities in humanoid robotics. Realizing large-area robotic skin requires integrating millions of tactile sensing elements across the body, creating major challenges in wiring complexity, power consumption, communication bandwidth, and real-time processing. Furthermore, conventional tactile systems continuously stream raw data to centralized processors, leading to scalability limitations and vulnerability to sensor failures, occlusions, and missing modalities. Similar limitations exist in current AI pipelines, where perception and control performance often degrade when sensory inputs become incomplete or unreliable. These challenges motivate a new generation of bio-inspired tactile systems that perform distributed sensing, local intelligence, and event-driven communication, transmitting only task-relevant information while maximizing robustness, efficiency, and scalability.

This PhD research aims to develop NeuromorphicTAC, a distributed tactile intelligence platform based on neuromorphic integrated circuits and edge AI architectures inspired by biological somatosensory pathways. The research will investigate how tactile sensing, signal processing, feature extraction, and sensor fusion can be embedded directly within distributed computing ASICs integrated into robotic skin. By processing tactile information locally and communicating sparse event-based representations rather than continuous raw data streams, the proposed architecture will enable ultra-low-power operation, low latency, fault tolerance, and scalable deployment across humanoid robots. The work will start with a comprehensive study of biological touch processing and state-of-the-art tactile systems, followed by the design, implementation, and validation of neuromorphic ASICs and distributed tactile-processing architectures. Validation will be performed at both component and system level using humanoid-relevant tasks, including object manipulation, contact-rich interaction, and whole-body tactile perception.

Key research challenges include achieving energy-efficient tactile intelligence for battery-powered robotic platforms, enabling real-time closed-loop control with millisecond latency, ensuring graceful degradation under sensor failures, and supporting seamless scaling from localized tactile patches to full-body robotic skin. The proposed NeuromorphicTAC architecture will be benchmarked against conventional centralized tactile-processing approaches, quantifying improvements in power efficiency, communication bandwidth, latency, robustness, and task performance. A unique opportunity lies in combining flexible and textile-integrated tactile sensors with neuromorphic edge intelligence, paving the way toward scalable, embodied AI systems capable of human-like touch perception.