Senior Software Engineer (Perception)

Role Overview

This position involves comprehensive responsibilities across the perception technology stack, including sensor calibration, multi-sensor data fusion, and vision models running both onboard robots and within data processing pipelines. The role demands a hands-on approach within a compact team, with the opportunity to tailor the scope based on your expertise and to advance towards technical leadership of the stack.

Primary Responsibilities

• Calibrating sensors such as depth cameras, 3D LiDAR, RGB cameras, Time-of-Flight (ToF), and sonar, including performing calibration in the field.
• Managing multi-sensor data processing and fusion, ensuring the stability and CPU efficiency of the point cloud pipeline.
• Developing and deploying onboard vision models for object detection, segmentation, and map-based zone proposals.
• Creating offboard vision models for automatic labeling, data mining, and generating synthetic datasets.
• Conducting research on emerging models including vision-language integration, monocular depth estimation, 3D reconstruction, and world modeling to support simulations and synthetic data creation.
• Overseeing the data pipeline and model lifecycle management, including dataset and model version control, experiment tracking, and gated over-the-air (OTA) updates of models.
• Implementing automated testing frameworks, benchmarking, and software-in-the-loop regression testing to enforce release quality standards.

Required Qualifications

• Bachelor's or Master's degree in Computer Science, Electrical Engineering, Robotics, or a closely related discipline.
• At least 4 years of practical experience in perception or computer vision applied to actual products.
• Proficiency in Python and PyTorch, supplemented by C++ for onboard software development.
• Demonstrated hands-on expertise in both sensor data processing (including calibration and handling point clouds from depth or LiDAR sensors) and vision model training using proprietary data; experience limited to only one of these areas is insufficient.
• Proven track record of deploying models on embedded hardware using frameworks such as OpenVINO, TensorRT, ONNX, RKNN, or Hailo; familiarity with ROS 2 is highly advantageous.