JOB ID 28251
Perception Algorithm Engineer
- メーカー, 自動車
- 製品エンジニア（ハード・ソフトウェア）
- 東京都
- 1000万円～2000万円
2024年4月に設立された日本のディープテック系スタートアップ企業。大型トラック向け自動運転システムの開発を専門とし、ドライバー不足や高齢化、物流需要の増加といった日本の物流業界が抱える課題の解決を目指している。高速道路を利用した幹線輸送に重点を置き、AIを活用した先進的な自動運転技術により、輸送の安全性と業務効率の向上、物流コストの削減を推進している。自動車メーカーや物流企業との連携を通じて、自動運転トラックの実用化と次世代の物流インフラの構築に取り組んでいる。
A Japanese deep-tech startup established in April 2024, specializing in the development of autonomous driving systems for heavy-duty trucks. The company aims to transform Japan’s logistics industry by addressing challenges such as driver shortages, an aging workforce, and increasing freight transportation demand. Focusing on autonomous highway transportation, it develops advanced AI-powered technologies designed to improve transportation safety, operational efficiency, and logistics sustainability. Through collaboration with automotive manufacturers and logistics companies, the company is working toward the commercialization of autonomous trucking solutions and the development of next-generation automated freight transportation infrastructure in Japan.
This role focuses on 2D / 3D object detection, online map perception, 3D occupancy prediction, and 3D Gaussian Splatting (3DGS) for driving-scene reconstruction and perception data generation. The position is responsible for model development, training, evaluation, and data-driven iteration using large-scale real-world driving data. You will develop perception algorithms that identify surrounding objects, estimate local road structure, and represent occupied and free space from onboard sensor observations to support downstream prediction and planning. Both full-time engineers and interns are welcome; interns will contribute to scoped projects with technical guidance.
募集要項
-
- 職種要約
- • Develop 2D and 3D object detection algorithms for road users and obstacles, including object classification, localization, dimensions, orientation, and velocity estimation where applicable.
• Develop online map perception algorithms to detect and reconstruct local road elements, including lane markings, road boundaries, crosswalks, and stop lines, with vectorized geometry and connectivity where required.
• Develop 3D occupancy and semantic occupancy algorithms to estimate occupied, free, and unknown space, including irregular obstacles beyond predefined object categories.
• Develop 3DGS-based driving-scene reconstruction and novel-view rendering pipelines using multi-view images, camera poses, and LiDAR data where available, with attention to static scene geometry and dynamic objects.
• Use reconstructed scenes to support perception data generation and scenario validation; evaluate geometric accuracy, rendering quality, view consistency, and downstream perception performance on real-world data.
• Build camera-based, LiDAR-based, or multimodal perception models, including BEV representations, sensor fusion, and temporal aggregation as appropriate for each task.
• Build and improve training and data pipelines using large-scale real-world driving data, with attention to sensor calibration, time synchronization, coordinate transforms, and annotation quality, including occupancy supervision generation, visibility masks, and treatment of unobserved space.
• Conduct training, debugging, ablation studies, and performance analysis to improve detection, map perception, and occupancy prediction across occlusion, long-range observations, adverse lighting, complex intersections, and other challenging scenarios.
• Design and maintain evaluation methods covering detection accuracy, localization quality, map geometry and connectivity, occupancy geometry and semantics, free-space precision and recall, temporal stability, and scenario-specific performance.
• Analyze false positives, missed detections, map reconstruction failures, and occupancy errors such as false-free predictions and missed thin obstacles; identify high-value data needs and drive targeted model and dataset improvements.
• Work with prediction, planning, engineering, and infrastructure teams on output interfaces, inference efficiency, deployment constraints, and integration readiness.
-
- 応募資格（必須）
- • Hands-on algorithm development experience in 2D object detection, 3D object detection, online map perception, 3D occupancy prediction, or 3DGS-based scene reconstruction, with a solid understanding of the related autonomous driving perception tasks.
• Strong programming ability in Python or C++, with practical experience in PyTorch or an equivalent deep learning framework and the ability to debug training and evaluation pipelines.
• Solid understanding of object detection fundamentals, including feature representations, detection heads, training objectives, sample assignment, post-processing, and evaluation metrics.
• Understanding of 3D geometry, camera projection, sensor calibration, coordinate transforms, and camera or point-cloud representations used in autonomous driving.
• Understanding of online map perception, including local road-element extraction, polyline or vector representations, and the distinction between geometric accuracy and connectivity correctness.
• Understanding of occupancy representations, voxel resolution and spatial extent, semantic labeling, visibility-aware supervision, and occupied / free / unknown space distinctions.
• For candidates focusing on 3DGS, understanding of multi-view geometry, Gaussian scene representations, differentiable rendering, reconstruction objectives, and camera pose or calibration errors.
• Ability to design controlled experiments, analyze long-tail failure cases, and translate data and metric findings into measurable improvements.
• Ability to maintain reproducible experiments and collaborate on reliable engineering integration.
Full-Time Engineers
• Practical experience developing and evaluating models using large-scale real-world or production driving data; experience limited to small academic datasets is not sufficient.
• Ability to independently own substantial model development, diagnose data and model failures, and deliver reproducible improvements.
• Experience collaborating with engineering teams on model interfaces, runtime constraints, and integration readiness.
-
- 勤務地
- 東京都
-
- 勤務スタイル
- 在宅勤務OK, フレックス勤務OK
-
- 給与
- 1000万円～2000万円
-
- こだわり条件
- 海外展開, 土日祝休み, 大手企業
Consultation
自動車・モビリティ業界における
転職やキャリア設計、
求人の選定の
ご相談を承ります。
