Senior Perception AI Engineer - Autonomous Vehicles

NVIDIA•Santa Clara, CA

About The Position

Build the online mapping system that converts data from multiple sensors into an accurate, continuously updated 3D model of the driving environment without relying on a prior map. The map represents lanes, road boundaries, crosswalks, intersection layouts, and traffic control information for use across the autonomous driving system. The role combines learned mapping—training DNNs to predict map elements from visual sensors, LiDAR, and radar—with geometry and C++ engineering to turn network outputs into stable, production-ready map elements. Models and algorithms are deployed on NVIDIA DRIVE Orin and Thor and tested in vehicles under real-time and safety constraints. The work supports NVIDIA L2++, L3, and L4 autonomous driving systems. The role also leads the ground-truth and data loop, including dataset planning, label and GT requirements, metric definition, failure analysis, and evaluation tooling. It works with labeling, data, and tooling partners and collaborates with perception, localization, planning, and infrastructure teams. The position contributes to technical direction and architecture across learned and geometric mapping components.

Requirements

MS or PhD, or equivalent experience, in Computer Science, Electrical Engineering, Robotics, or a related field.
At least 5 years of industry or research experience in autonomous driving, robotics, or computer vision.
Strong foundation in 3D geometry, including camera models, projective geometry, coordinate transforms, optimization, and curve or surface fitting.
Strong C++ skills and a track record of designing and shipping fast, reliable, maintainable production-quality algorithm code.
Hands-on experience training deep neural networks in PyTorch, including model design, training pipelines, debugging, and performance analysis.
Experience with data-driven development, including building datasets, defining labels and ground truth, choosing metrics, and improving models through failure analysis.
Excellent problem-solving and communication skills, and the ability to drive technical direction across teams.

Nice To Haves

Direct experience with DNN-based online or HD map generation, BEV perception, vectorized map prediction, or lane-topology learning, or equivalent BEV-perception experience.
Experience with intersection understanding, including lane connectivity and topology reasoning in complex junctions, or traffic-light detection and lane association.
Experience combining learned models with geometric post-processing in a production autonomous-driving system.
Publications or open-source contributions in 3D vision, mapping, or autonomous driving.
Background in sensor fusion, multi-object tracking, or temporal fusion for map elements.
Experience deploying DNNs and algorithms on embedded GPUs using CUDA or TensorRT, ideally on DRIVE Orin or Thor, under real-time constraints.

Responsibilities

Design, train, and improve PyTorch models that predict online map elements such as lanes, road boundaries, crosswalks, road markings, and topology from camera, LiDAR, and/or radar data using BEV and vectorized-map representations.
Develop models for intersection topology, including lane connectivity, turn paths, merges, and splits, and for traffic-light detection, state estimation, and association with controlled lanes.
Write production C++ for geometry-based post-processing, including curve and surface fitting, association, multi-frame fusion and tracking, topology reasoning, and consistency checks on DNN outputs.
Apply multi-view geometry, calibration, coordinate transforms, and sensor fusion to create accurate, stable 3D maps over time.
Optimize and deploy models and algorithms on NVIDIA DRIVE Orin and Thor under real-time and safety constraints, and get them running in vehicles.
Plan datasets and define ground-truth and labeling requirements; work with labeling, data, and tooling partners to produce high-quality ground truth.
Mine data, select metrics, and analyze failure cases to drive improvements.
Define critical metrics and build evaluation tooling to measure quality gains against large-scale real-world data.
Partner with perception, localization, planning, and infrastructure teams to integrate online mapping into the autonomous-driving stack across L2++, L3, and L4 programs.
Contribute to technical direction and architecture decisions across the learned and geometric components of the mapping system.