Perception Algorithm Engineer

About the Role

This position focuses on cutting‑edge development of World Models, Foundation Models, and multimodal Bird’s‑Eye‑View (BEV) algorithms for an autonomous driving perception stack. You will drive end‑to‑end model development, large‑scale training experiments, sophisticated evaluation frameworks, and data‑driven iteration using massive volumes of real‑world heavy‑trucking data.

Key Responsibilities

Algorithm Development: Architect and implement World Models, Foundation Models, and multimodal BEV perception frameworks tailored for autonomous driving; design advanced scene representation, temporal modeling, multimodal fusion, and perception‑oriented prediction solutions.
Pipeline Construction: Build, scale, and iterate neural‑network architectures and training pipelines that ingest heterogeneous real‑world data (camera, LiDAR, radar, vehicle‑state signals).
Experimentation & Optimization: Lead full‑lifecycle model training, hands‑on debugging, ablation studies, and deep performance analysis to improve perception quality across complex, edge‑case scenarios.
Evaluation Frameworks: Design and maintain multi‑dimensional evaluation paradigms, including scenario‑based metrics, long‑tail edge‑case analysis, generalization assessments, and regression tracking tools.
Data‑Driven Iteration: Isolate and analyze model failure modes from fleet data; translate insights into actionable data‑requirements for active learning, data curation, and targeted dataset improvements.
Cross‑Functional Collaboration: Work closely with core engineering and infrastructure teams to validate model feasibility, optimize distributed training efficiency, respect real‑time inference constraints, and ensure seamless production integration.

Essential Qualifications

Hands‑on experience in at least one of the following domains: World Models, Foundation Models, multimodal BEV perception, large‑scale representation learning, generative models, or end‑to‑end autonomous driving models.
Practical experience training, adjusting, and deploying models with large‑scale, real‑world production driving data (small academic benchmarks alone are insufficient).
Strong programming proficiency in Python or C++, with the ability to write clean, maintainable, production‑ready research code and efficiently debug complex distributed training pipelines.
Solid understanding of modern autonomous‑driving perception components such as BEV perception, multimodal sensor fusion, temporal modeling, 3D object detection, multi‑object tracking, semantic occupancy, or scene‑level representation learning.
Analytical mindset: demonstrated ability to diagnose and unpack model behavior using statistical data, granular metrics, and real‑world failure cases rather than focusing solely on academic benchmark scores.
Fluent English communication; Japanese language skills are a plus.

Preferred (Nice‑to‑Have) Skills

Safety‑Critical Embodied AI: experience applying large‑scale foundation or world models to autonomous driving fleets, robotics, commercial vehicles, or other safety‑critical embodied AI systems.
Deployment & Acceleration: familiarity with hardware inference latency targets, memory footprint optimization, model compression, quantization (INT8/FP16), and acceleration frameworks such as TensorRT.
Infrastructure & Simulation: experience with distributed multi‑node training clusters, advanced data curation strategies, synthetic data generation, simulation pipelines, or closed‑loop evaluation loops.
Research Track Record: relevant publications or open‑source contributions in computer vision, machine learning, autonomous driving, robotics, or foundation model research, combined with strong practical engineering experience.