摘要
This demo showcases a system that can detect obstacles and control a character in the simulator to avoid collisions in real-time. This system consists of the following components: a low-power CMOS image sensor that provides high-resolution images, a lightweight spiking neural network that uses monocular inertial-visual input to sense the depth of obstacles, and a spiking neural network accelerator that speeds up the heavy computation of the neural network. The entire system can run at 25 FPS and demonstrates the capability to avoid obstacles quickly and accurately.