Event-based vision sensors achieve up to threeorders of magnitude better speed vs. power consumption tradeoff in high-speed control of UAVs compared to conventionalimage sensors. Event-based cameras produce a sparse streamof events that can be processed more efficiently and with alower latency than images, enabling ultra-fast vision-drivencontrol. Here, we explore how an event-based vision algorithmcan be implemented as a spiking neuronal network on aneuromorphic chip and used in a drone controller. We show howseamless integration of event-based perception on chip leadsto even faster control rates and lower latency. In addition, wedemonstrate how online adaptation of the SNN controller can berealised using on-chip learning. Our spiking neuronal networkon chip is the first example of a neuromorphic vision-basedcontroller solving a high-speed UAV control task. The excellentscalability of processing in neuromorphic hardware opens thepossibility to solve more challenging visual tasks in the futureand integrate visual perception in fast control loops.

No abstract