Self healable neuromorphic memtransistor elements for decentralized sensory signal processing in robotics

Area of Science:

• Robotics and Neuromorphic Engineering
• Materials Science and Engineering

Background:

• Current robotic skin sensory processing relies on centralized systems, leading to data transfer issues, latency, and reduced robustness.
• A decentralized approach with embedded intelligence in sensing nodes offers a more efficient and resilient solution for robotic sensory information processing.

Purpose of the Study:

• To develop a decentralized neuromorphic system for robotic skins that mimics pain perception (nociception) and tactile association.
• To enable robots to trigger an escape reflex based on sensory input, improving safety and interaction.

Main Methods:

• Utilized self-healable materials and memristors configured as gated-threshold and memristive switches.
• Implemented a unique neuromorphic methodology for localized information extraction and associative learning within sensing nodes.
• Developed in-memory edge computing capabilities with minimal hardware and wiring.

Main Results:

• Demonstrated the successful implementation of neuromorphic nociceptors and spiking local associative learning.
• Achieved in-memory edge computing with reduced hardware complexity and enhanced fault tolerance.
• Successfully triggered an escape reflex in a sensorized robotic arm based on simulated nociception and tactile input.

Conclusions:

• The proposed decentralized neuromorphic system offers a robust and efficient approach to sensory information processing in robotic skins.
• Memristor-based neuromorphic nociceptors and edge computing significantly improve robotic system resilience and reduce data transfer burdens.
• This methodology paves the way for more sophisticated and human-like sensory perception and phản ứng in robots.