Machine Learning Engineer (Egocentric 3D Human Pose)
关于角色 我们正在寻找机器学习工程师加入我们的核心研究和开发团队，专注于从自我中心 （第一人）视频中恢复精确的3D 人类身体和手动。人体示范数据是机器人学习的基础，其质量取决于准确重建人类运动。在这个职位上，你将开发模型和生产管道，这些模型可以转化头部和身上的摄像头流，包括宽光伏，立体音响，运动模糊和高度自闭的视频到测量准确，时间一致的3D 表示来进行机器人的政策训练和人对机器人运动重定向。你将在整个感知堆中工作，包括摄像头校准，数据注释，模型培训，评估和大型部署。这项职位适合具有计算机视觉和深度学习方面的强大的专业知识的工程师，他们喜欢解决现实世界复杂的感知问题。责任 ·开发最先进的3D 体型和手动姿势估计模型用于单向和立体摄像机系统使用自我中心的视频。·构建2D/3D 键点估算，SMPL/SMPL-X,MANO 以及全身运动重组模型。·解决具有挑战性的自我中心视力问题，包括严重的自闭，运动模糊，滚动离子器文物，截肢，极端视角和手对象互动。·设计和维护摄像头几何和校准管道，包括鱼眼和宽光相机模型，立体校准，三角形和协调框架配合。·通过过，动态限制，多视觉融合和多模式传感器集成提高时间一致性和物理可靠性。·为大型人类运动数据集构建可扩展的注释，评估和质量保证管道。· 优化大型模型培训和高吞吐量推断管道用于生产环境。· 与机器人工程师密切合作，将重建的人类运动转换为高质量的机器人培训数据。· 贡献系统架构，工程最佳实践以及感知平台的长期发展。最低资格。·计算机科学，机器学习，计算机视觉，机器人技术或相关领域的学士，士或博士学位。· 3年以上建立和部署机器学习系统经验。· 用3D 人类估计来进行实践经验，
查看英文原文
About the Role
We are looking for a Machine Learning Engineer to join our core research and development team focused on recovering accurate 3D human body and hand motion from egocentric (first-person) video.
Human demonstration data is the foundation of robot learning, and its quality depends on accurately reconstructing human motion. In this role, you will develop models and production pipelines that transform head-mounted and body-mounted camera streams—including wide-FOV, stereo, motion-blurred, and heavily self-occluded video—into metrically accurate, temporally consistent 3D pose representations for robot policy training and human-to-robot motion retargeting.
You will work across the entire perception stack, including camera calibration, data annotation, model training, evaluation, and large-scale deployment. This role is ideal for engineers with strong expertise in both computer vision and deep learning who enjoy solving challenging real-world perception problems.
Responsibilities
- Develop state-of-the-art 3D body and hand pose estimation models for egocentric video using monocular and stereo camera systems.
- Build models for 2D/3D keypoint estimation, SMPL/SMPL-X, MANO, and full-body motion reconstruction.
- Address challenging egocentric vision problems, including severe self-occlusion, motion blur, rolling shutter artifacts, truncated limbs, extreme viewpoints, and hand-object interaction.
- Design and maintain camera geometry and calibration pipelines, including fisheye and wide-FOV camera models, stereo calibration, triangulation, and coordinate frame alignment.
- Improve temporal consistency and physical plausibility using filtering, kinematic constraints, multi-view fusion, and multi-modal sensor integration.
- Build scalable annotation, evaluation, and quality assurance pipelines for large-scale human motion datasets.
- Optimize large-scale model training and high-throughput inference pipelines for production environments.
- Collaborate closely with robotics engineers to convert reconstructed human motion into high-quality robot training data.
- Contribute to system architecture, engineering best practices, and the long-term evolution of the perception platform.
Minimum Qualifications
- Bachelor's, Master's, or PhD in Computer Science, Machine Learning, Computer Vision, Robotics, or a related field.
- 3+ years of experience building and deploying machine learning systems.
- Hands-on experience with 3D human pose estimation, hand pose estimation, or human motion tracking from video.
- Strong understanding of multi-view geometry, camera calibration, triangulation, coordinate transformations, and projection models.
- Strong Python programming skills and proficiency with PyTorch or TensorFlow.
- Solid knowledge of modern deep learning techniques, model training, evaluation, and production ML workflows.
- Strong analytical and problem-solving skills with the ability to thrive in a fast-paced collaborative environment.
Preferred Qualifications
- Experience with egocentric perception systems, AR/VR headsets, smart glasses, or wearable capture rigs.
- Expertise with SMPL, SMPL-X, MANO, inverse kinematics, markerless motion capture, or hand-object pose estimation.
- Familiarity with egocentric vision datasets and benchmarks.
- Experience with modern video and 3D learning architectures, including Video Transformers, diffusion-based motion models, or 3D CNNs.
- Experience building multi-camera capture systems, synchronization, and calibration infrastructure.
- Experience with human-to-robot motion retargeting, teleoperation, imitation learning, or dexterous manipulation.
- Experience developing annotation tools, active learning pipelines, or large-scale data quality systems.
- Publications at leading conferences such as CVPR, ICCV, ECCV, NeurIPS, SIGGRAPH, or 3DV, open-source contributions, or demonstrated impact in applied AI systems.
What We Offer
- Competitive salary and equity package
- Comprehensive medical, dental, and vision insurance
- 401(k) retirement plan
- Generous paid time off and company holidays
- Paid sick leave
- Opportunity to work alongside leading researchers and engineers in robotics, computer vision, and AI
- High-impact role building cutting-edge perception systems for next-generation robotics
Originally posted on Himalayas