Machine Learning Engineer

About This Role

Machine Learning Engineers at Lucid develop scalable auto-labeling technologies to transform fleet-collected sensor data into high-quality training datasets with minimal human intervention. This role sits at the intersection of machine learning, computer vision, spatial AI, and large-scale data infrastructure, working on perception models, 3D scene reconstruction, foundation models, and automated data-generation systems.

Responsibilities

• Design and deploy production-scale auto-labeling solutions for 2D and 3D object detection, semantic and panoptic segmentation, lane and road-structure extraction, multi-object tracking, occupancy, scene understanding, and traffic-participant behavior
• Develop automated label validation, confidence scoring, consistency checks, and quality-monitoring frameworks
• Build active-learning and human-in-the-loop workflows that improve label quality while reducing manual annotation effort and turnaround time
• Create teacher-student and model-assisted labeling workflows for large-scale camera and LiDAR datasets
• Research, train, and evaluate camera-based and LiDAR-based detection, segmentation, tracking, depth-estimation, and occupancy models
• Develop multimodal fusion models combining camera, LiDAR, radar, GNSS/IMU, and map context
• Improve temporal consistency, calibration robustness, and long-tail performance across diverse driving environments and edge cases
• Build end-to-end ML workflows from raw sensor ingestion and preprocessing through annotation, quality control, and dataset delivery
• Develop 3D Gaussian Splatting (3DGS) and related reconstruction pipelines for large-scale driving scenes collected from fleet vehicles
• Use reconstructed scenes for label propagation, synthetic annotation generation, annotation validation, rare-case augmentation, and simulation
• Create scene-centric representations that improve geometric accuracy and consistency across sensors, viewpoints, and time
• Apply Vision-Language Models (VLMs), Vision-Language-Action (VLA) models, and multimodal foundation models to scene interpretation, label generation, verification, and edge-case discovery
• Develop natural-language and agent-based workflows that help engineers search, inspect, curate, and validate large-scale datasets
• Evaluate state-of-the-art research and convert promising methods into reliable production capabilities
• Build scalable training and inference pipelines for fleet-scale datasets and distributed GPU infrastructure
• Optimize model performance, throughput, reliability, monitoring, and reproducibility across cloud and production environments
• Collaborate with perception, mapping, navigation, simulation, hardware, logging, and data-platform teams to deploy end-to-end solutions
• Define and track metrics for labeling accuracy, coverage, throughput, cost, and operational efficiency