Falcon Vision-Inspired Ultrafast Traffic Obstacle Avoidance Based on 2D Edge-Rich van der Waals Heterostructures.

Abstract

Ultrafast and reliable visual perception is essential for obstacle avoidance in autonomous driving, where split-second decisions must be made in complex, high-speed environments, yet remains constrained by the limited temporal resolution and processing latency of conventional devices. Here, inspired by the exceptional temporal resolution of falcon vision systems (>150 Hz), we develop a neuromorphic vision sensor capable of ultrafast, edge-selective perception for dynamic traffic scenarios. The sensor leverages vertically stacked, edge-rich SnS₂/MoS₂ van der Waals heterostructures, in which a high density of atomic-scale interfaces and defective edges enables enhanced light-matter interactions and rapid carrier dynamics. These structural advantages endow the Falcon Vision Sensor (FVS) with synaptic plasticity (PPF = 201%, LTP = 1300s), high refresh rate (250 Hz), and intrinsic erasure behaviors, closely mimicking the temporal precision and motion discrimination features of falcon vision. When the synaptic devices are integrated with computing modules, the system achieves real-time obstacle detection, along with a directional motion recognition accuracy of 98.89%. This work demonstrates a robust biologically inspired visual intelligence, offering a compact, low-latency solution for next-generation autonomous vehicles and edge AI applications requiring rapid environmental responsiveness.