Perovskite retinomorphic image sensor for embodied intelligent vision.

Area of Science:

• Robotics and Artificial Intelligence
• Materials Science and Engineering
• Neuroscience and Bio-inspired Computing

Background:

• Exoskeleton robots require advanced vision systems for enhanced human assistance.
• Current retinomorphic devices lack sensitivity, reconfigurability, and large-scale integration.
• Developing efficient edge-processing algorithms is crucial for realizing retinomorphic hardware.

Purpose of the Study:

• To develop a retinomorphic hardware prototype with embodied intelligent vision functionalities.
• To create a system capable of real-time environmental perception and response.
• To advance the capabilities of exoskeleton robots for assisting physically disabled individuals.

Main Methods:

• Developed a 4096-pixel perovskite image sensor array as the core module.
• Employed a one photodetector-one transistor geometry mimicking retinal circuits.
• Implemented adaptive imaging with contrast enhancement and 1D feature extraction algorithms.

Main Results:

• Achieved ~620% contrast enhancement under low-light conditions (10 microwatts/cm²).
• Demonstrated broadband, ultrahigh, and reconfigurable photoresponsivities.
• Successfully decomposed visual scenarios into encoded spatiotemporal information.

Conclusions:

• The retinomorphic system provides embodied intelligence with adaptive imaging, in situ processing, and decision-making.
• This technology holds significant potential for autonomous robot applications, particularly in assistive robotics.
• The perovskite-based sensor array offers a scalable and effective solution for advanced robotic vision.