A-RoI: Adaptive Region of Interest to Enhance Obstacle Awareness for Autonomous Driving

Autonomous vehicles navigating dynamic environments demand adaptive perception and planning for safe operation. Crucially, navigational decisions rely on understanding the position and behaviour of surrounding objects. Traditional, static Regions of Interest (RoIs) for obstacle detection prove inadequate in meeting these dynamic and context-dependent requirements. This paper presents Adaptive Region of Interest (A-RoI), a novel technique that dynamically updates the RoI on a LiDAR HD map based on the vehicle's real-time position and planned trajectory. Integrating obstacle positions with the A-RoI within a Bird's Eye View (BEV) framework enables effective obstacle perception, significantly enhancing navigational decision-making. Real-time validation on a sensorequipped autonomous vehicle demonstrates the effectiveness of the proposed approach. Future research will explore enhancing A-RoI's adaptability through multi-modal sensor fusion and incorporating learning-based methodologies for further improvements in perception and decision-making capabilities.