Neuromorphic Event-Driven Spike Routing for Low-Latency Visual Odometry
Conventional frame-based sensors exhibit high motion blur and prohibitive energy consumption in high-velocity drone flight. Here, we present a spiking neural network (SNN) topology integrated with neuromorphic silicon event cameras. By executing temporal spike contrast backpropagation on an asynchronous crossbar array, our model tracks 6-DOF trajectory poses at 10,000 frames-per-second equivalent temporal resolution with only 14mW power consumption.