Edge-AI-Based Decision Systems for Autonomous Ground Robots

Abstract

Autonomous Ground Robots (AGRs) operating in dynamic, time-critical environments require immediate situational awareness and microsecond-level behavioral adaptation. Conventional cloud-coupled robotic frameworks suffer from unpredictable latency spikes, network bandwidth saturation, and vulnerabilities to communication dropouts. This paper introduces a robust, fully decentralized Edge-AI architecture designed explicitly for real-time navigation, dynamic hazard detection, and kinematic control optimization in AGRs. By deploying highly optimized convolutional neural networks and deep reinforcement learning algorithms directly onto hardware-constrained onboard accelerators, the proposed system enables localized intelligence without remote server dependency. Empirical validations conducted across heterogeneous simulated and physical testbeds demonstrate that the decentralized Edge-AI approach reduces localized control loops down to the sub-millisecond level while maintaining minimal power consumption profiles. The findings confirm that localized model execution provides the computational reliability and functional safety mandatory for complex, industrial, and disaster-recovery field deployments.