Real-Time Object Detection as a Service for UGVs Using Edge Cloud

Abstract

Autonomous navigation has made significant strides in recent years, finding successful deployment in controlled environments. This achievement has been facilitated by the increased computational power and machine learning techniques. Nevertheless, overcoming numerous challenges is crucial for its widespread adoption across all environments. One notable obstacle involves the provision of dependable, low-latency, and cost-effective data processing solutions for compute-intensive applications. To tackle this challenge, this demo investigates the potential for offloading compute-intensive tasks from a UGV to a nearby edge cloud and characterize the performance in terms of latency and throughput. By doing so, compute-heavy workloads on a UGV are replaced by simple API calls to edge cloud-based services deployed. It also keeps the UGV system design simple, reduces hardware costs, and saves power consumption. This approach offers significant benefits for autonomous vehicles in controlled environments such as campus shuttles, agricultural rovers, and warehouse rovers.