Real-Time Road, Lane and LiDAR Segmentation on Embedded Hardware
Light perception networks that find vehicles, drivable area, lanes and LiDAR classes fast enough for embedded boards in vehicles.
Tier Grades
Tier A: Frontier AI labs
stretch
Frontier labs are not observed here; the actors are universities and chip vendors.
Tier B: Indonesian AI labs
feasible
YOLOP runs at 23 FPS on a Jetson TX2 and trains on the public BDD100K dataset.
Tier C: Institution units such as ITB
feasible
YOLOP runs at 23 FPS on a Jetson TX2 and trains on the public BDD100K dataset.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
A 23 FPS model on an embedded board fits a fleet safety or site vehicle proof of concept.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeYOLOP: 23 FPS on an embedded Jetson TX2.
- DataYOLOP: BDD100K driving dataset; SalsaNext: LiDAR point clouds.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 7 authors on the YOLOP paper (arXiv 2108.11250).
- Time horizonMonths.
- Shelf-life signalsHybridNets followed YOLOP within a year on the same three tasks.
- Pipeline roleUniversities publish light networks; Tier B and Tier C retrain them on local roads; Tier D fits them to fleet safety and site vehicles.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
Links
R&D TopicReal-Time Road, Lane and LiDAR Segmentation on Embedded Hardware
Models
YOLOPHuazhong University of Science and Technology, drivingHybridNetsdriving
SalsaNextHalmstad University, driving
MinkUNetStanford University, vision+5 more
YOLOPHuazhong University of Science and Technology, drivingHybridNetsdrivingSalsaNextHalmstad University, drivingMinkUNetStanford University, vision+5 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 9
SPVCNNTsinghua University, Massachusetts Institute of Technology, visionproduced by
Cylinder3DThe Chinese University of Hong Kong, visionproduced byMinkUNetStanford University, visionproduced bySalsaNextHalmstad University, Volvo Group, drivingexemplar
Intel road and ADAS segmentationIntel, drivingproduced by
LaneNetKU Leuven, drivingproduced byYOLOPHuazhong University of Science and Technology, drivingexemplarHybridNetsdrivingexemplarRangeNet++drivingproduced byPapers
- SqueezeSeg: Convolutional Neural Nets with Recurrent CRF for Real-Time Road-Object Segmentation from 3D LiDAR Point CloudBichen Wu et al., arXiv preprint, 2017 founding
- Point Transformer V3: Simpler, Faster, StrongerXiaoyang Wu et al., CVPR 2024 (per arXiv comment), 2023 state of the art
- 2DPASS: 2D Priors Assisted Semantic Segmentation on LiDAR Point CloudsXu Yan et al., ECCV2022 (per arXiv comment), 2022 state of the art
- PolarNet: An Improved Grid Representation for Online LiDAR Point Clouds Semantic SegmentationYang Zhang et al., CVPR 2020 (per arXiv comment), 2020 state of the art
- Cylindrical and Asymmetrical 3D Convolution Networks for LiDAR SegmentationXinge Zhu et al., arXiv preprint, 2020 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2108.11250 https://arxiv.org/abs/2108.11250
- arxiv.org/abs/2203.09035 https://arxiv.org/abs/2203.09035
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Real-Time Road, Lane and LiDAR Segmentation on Embedded Hardware", https://aurion.semai.id/topics/real-time-road-and-lidar-segmentation/, records of October 2026.