Vision Algorithm Engineer

Role Overview

This position focuses on developing and deploying vision perception algorithms for photovoltaic (PV) cleaning robots. The role includes the creation of advanced vision modules for object detection, navigation, obstacle classification, and surface contaminant analysis on PV modules. It involves end-to-end management of vision model pipelines applied in industrial environments, including data handling, annotation, model training, optimization, and deployment on embedded robotic platforms.

Key Responsibilities

• Develop vision perception modules including object detection, instance segmentation of PV modules, recognizing navigation routes, detecting and classifying obstacles such as cables, posts, and weeds, and analyzing surface contaminants like dust, bird droppings, and snow accumulation.
• Manage the entire lifecycle of vision models encompassing data collection, cleansing, annotation strategy design, model selection, training, optimization, and deployment.
• Deploy cutting-edge 2D/3D vision algorithms and deep learning models (e.g., YOLO series, Mask R-CNN, Segment Anything Model) on robotic embedded platforms such as the Jetson series and balance performance between speed and accuracy through tuning.
• Explore sensor fusion strategies integrating visual data with inputs from LiDAR and IMU sensors to enhance perception for robot localization, navigation, and decision-making.
• Stay current with advancements in computer vision and deep learning fields such as Transformers, Vision-Language Models, and domain adaptation, assessing their potential to enrich robot environmental understanding and intelligence.
• Collaborate with algorithm, hardware, and mechanical teams during requirement review and technical solution formulation.
• Engage in on-site debugging to overcome vision-related challenges posed by complex lighting conditions, seasonal variations, and adverse weather (rain, fog), improving environmental adaptability of algorithms.

Candidate Qualifications

• Master's degree or higher in Computer Science, Artificial Intelligence, Automation, Electronic Engineering, or related disciplines.
• Minimum of 2 years professional experience demonstrating comprehensive knowledge of computer vision fundamentals and deep learning architectures, including CNN, RNN, and Transformer.
• Proficient in object detection, instance and semantic segmentation with practical experience in model training and optimization, using frameworks such as Faster R-CNN, YOLO, Mask R-CNN, and DeepLab.
• Strong commitment to responsibilities, a collaborative team spirit, and effective communication abilities.