LiDAR Perception and Real-Time Decision-Making for Automated Driving: From Point Clouds to Deployed Vehicles

LiDAR gives an automated vehicle the geometry of its surroundings, and it does so at a cost the rest of the system has to absorb: dense point clouds arriving many times a second, to be segmented, classified, tracked and turned into a decision before the vehicle has travelled a few metres. Most published progress sits at one end of that chain or the other. Perception papers report detection accuracy on curated benchmarks; planning and decision papers assume perception that is clean, complete and instant. In a deployed vehicle neither holds. The sensor itself is also changing: mechanical spinning units now sit alongside solid-state and MEMS scanners, FMCW devices and single-photon avalanche diode (SPAD) LiDARs, whose range, resolution and noise behaviour alter what the perception stack receives. Occlusion, weather, reflectivity, sensor placement and compute budget shape what the decision layer receives, and the latency of the whole pipeline decides whether a correct decision arrives in time. Infrastructure-mounted LiDAR and cooperative perception extend the horizon further, and add questions about synchronisation, trust and fallback.This Research Topic treats perception and decision-making as one real-time chain rather than two separate literatures. We seek work that reports what an algorithm costs as well as what it achieves — latency, compute, degradation under adverse conditions — and that shows the consequence for the driving decision downstream. Three strands are of particular interest. First, LiDAR perception itself: point cloud segmentation, detection and tracking, fusion with camera and radar, and the effect of sensor type, layout and resolution on what can be perceived. Second, the decision layer: manoeuvre selection, planning and control that act on imperfect perception, including explicit handling of uncertainty, occlusion and sensor failure. Third, the vehicle in its setting: infrastructure-mounted and cooperative LiDAR that extend the perceptual horizon through V2I and V2V, and the architectures that keep onboard and off-board information consistent. We prize evidence from real vehicles, test tracks and public-road pilots, including automated shuttles and confined-area services, over benchmark scores alone.We welcome contributions on, but not limited to: • LiDAR point cloud processing, 3D detection, segmentation and tracking • LiDAR sensor technologies and their consequences for perception, including mechanical, solid-state and MEMS scanning, FMCW and single-photon avalanche diode (SPAD) LiDAR • Multi-sensor fusion and calibration • Sensor placement, coverage and sight-distance analysis • Roadside and infrastructure lidar for traffic monitoring and for supporting automated vehicles • Cooperative and collective perception over v2i and v2v • Real-time decision-making, manoeuvre planning and control under perceptual uncertainty • Software architectures, middleware and computational budgets for the perception-to-decision chain • Datasets, annotation methods and benchmarks, including scalable and reproducible data collection • Validation in automated shuttles, valet parking, confined areas and public-road pilots • Acceptance and perceived risk of the decisions these systems make in real driving Original Research is the primary article type; Methods, Review, Brief Research Report and Perspective articles are also welcome. Shared datasets and released code are strongly encouraged. Topic editors: Felipe Jimenez, Jose Eugenio Naranjo Source: https://www.frontiersin.org/research-topics/86545/lidar-perception-and-real-time-decision-making-for-automated-driving-from-point-clouds-to-deployed-vehicles