Offloading Real-Time Collision Avoidance to the Edge-Cloud: Improving Energy Efficiency in Autonomous Systems

Autonomous mobile robots operating in structured environments rely on vision-based navigation to follow paths and avoid obstacles. These robots typically run computationally intensive machine learning models onboard, leading to high resource utilization, increased power consumption, and challenges in battery management.To address these constraints, this paper explores an edge-cloud offloading approach where resource heavy machine learning models are deployed as API based services on GPU equipped edge-cloud. By offloading computation from onboard systems to edge-cloud infrastructure, the system reduces processing load, improves energy efficiency, and maintain real-time decision-making with manageable latency. Real-time experiments using a Jetacker robot on a real-time road map scenario demonstrate a 47% reduction in RAM utilization, a 10% decrease in CPU usage, and near-zero GPU consumption onboard. As a result, the system’s overall power consumption is significantly reduced, from roughly 6660 mW to 3400 mW, which improves energy economy and extends the battery life for autonomous navigation.