Machine Learning Engineer: Perception Analytics

Here’s the reformatted HTML for the **Machine Learning Engineer: Perception** posting (as an example of a substantive role; apply the same pattern to other roles as needed):

San Francisco, CA or New York, NY · Full time · Hybrid

About the role

Build the perception systems that allow our robots to understand and interact with the physical world in real time.

Responsibilities

• Design, train, and deploy deep-learning models for 3D object detection, segmentation, and tracking using camera, LiDAR, and radar data.
• Develop novel architectures that fuse multi-modal sensor inputs and run efficiently on edge hardware.
• Curate and label large-scale datasets, implement active-learning pipelines, and establish rigorous evaluation metrics.
• Collaborate with robotics engineers to integrate perception outputs into planning and control stacks.
• Optimize models for latency, memory footprint, and power consumption without sacrificing accuracy.
• Publish research, file patents, and present findings at top-tier conferences (CVPR, ICCV, ICRA, etc.).

Requirements

• Master’s or PhD in Computer Science, Robotics, or a related field, or equivalent industry experience.
• 3+ years of hands-on experience building and shipping perception models for robotics, autonomous vehicles, or AR/VR.
• Proficiency in Python, PyTorch/TensorFlow, and modern deep-learning frameworks.
• Experience with sensor fusion, calibration, and geometric computer vision (SLAM, SfM, multi-view geometry).
• Strong software-engineering skills: version control, testing, CI/CD, and deployment on embedded platforms.
• Track record of solving real-world perception challenges in dynamic, unstructured environments.

Skills

• 3D deep learning (PointNet++, Transformer-based architectures)
• Multi-modal fusion (camera + LiDAR + radar)
• Model quantization, pruning, and deployment on NVIDIA Jetson / Qualcomm platforms
• ROS/ROS2, Docker, and real-time systems
• Data annotation tools (Labelbox, CVAT, custom pipelines)
• C++ for performance-critical components