Data Algorithm Engineer — Embodied Intelligence Human Video Direction

Build scalable human first-person video data systems for embodied intelligence foundation models.

What You Will Do

Human Ego Video Data System Construction

Build a human first-person video data system for training embodied intelligence foundation models. Around core scenarios such as human operation behavior, hand interaction, body motion, object manipulation, and scene understanding, construct high-quality, scalable, and continuously iterable embodied intelligence data assets.

Large-Scale Ego Video Data Management

Responsible for cleaning, filtering, deduplication, segmentation, retrieval, storage, version management, and distribution of large-scale human ego video data. Support efficient management and training usage of TB-to-PB-level multimodal video data.

Ego Video Annotation System Design

Design automated/semi-automated annotation specifications and production workflows for human first-person videos, including hand trajectories, body poses, action phases, object interactions, operation intent, task steps, key frames, failure behaviors, and long-horizon operation processes.

Hand Tracing and Body Tracing Algorithm Optimization

Develop and optimize algorithms for hand tracking, body pose estimation, action trajectory extraction, and hand-object interaction modeling in human ego videos, improving the automated extraction quality of hand, body, and action trajectory data.

Automated Annotation and Data Production Pipeline

Build large-scale automated/semi-automated annotation pipelines for human ego videos. Combine vision models, multimodal models, prompt engineering, rule engines, and human quality inspection workflows to improve annotation efficiency and accuracy.

Data Quality Evaluation and Closed-Loop Iteration

Establish an Ego video data quality evaluation system. Continuously optimize data filtering, annotation, and sampling strategies, and improve data quality based on model training results and failure cases.

Data Processing Inference Deployment and Performance Tuning

Responsible for inference deployment and performance optimization of video data processing, tracking algorithms, automatic annotation models, and data quality inspection models, ensuring efficient operation of large-scale data production pipelines.

Collaboration with Embodied Intelligence Model Training

Work closely with algorithm teams in embodied intelligence, VLA foundation models, multimodal foundation models, and robot learning, focusing on imitation learning, action prediction, world models, video-action modeling, and policy learning.

What We Expect

Academic Background

Background in computer vision, artificial intelligence, robotics, data science, computer science, automation, electronic information, mathematics, or related fields. Bachelor’s degree or above, with a solid foundation in vision algorithms, data engineering, and machine learning.

Ego Video Data Experience

Familiar with the characteristics of human first-person video data and understands the value of Ego videos in embodied intelligence, robot learning, imitation learning, action prediction, and vision-language-action modeling. Experience with Ego videos, human operation videos, robot operation videos, or multimodal video data processing is preferred.

Large-Scale Data Management Ability

Experience managing large-scale video or multimodal data. Familiar with the full process of data collection, cleaning, deduplication, segmentation, indexing, version management, quality evaluation, data distribution, and training set construction. Able to support stable production and iteration of TB-to-PB-level data.

Video Annotation and Data Production Experience

Experience designing or implementing large-scale video data annotation systems. Familiar with annotation tasks such as keypoints, trajectories, action phases, object interaction, task steps, temporal boundaries, and semantic labels. Able to formulate annotation specifications, quality inspection standards, and delivery acceptance workflows.

Hand / Body Tracking Algorithm Ability

Familiar with hand tracking, human pose estimation, keypoint detection, object detection, segmentation, optical flow, trajectory modeling, and temporal modeling. Experience in hand tracing, body tracing, hand-object interaction, pose estimation, or motion tracking projects is preferred.

Programming and Algorithm Implementation Ability

Proficient in Python and familiar with PyTorch or other mainstream deep learning frameworks. Skilled in using OpenCV, NumPy, Pandas, and other data/video processing tools. Strong coding ability, able to independently develop algorithm experiments, data processing scripts, automated annotation tools, and evaluation tools.

Inference Deployment and Performance Optimization Ability

Experience in model inference deployment and data processing performance tuning. Familiar with batch inference, distributed inference, GPU inference optimization, memory optimization, multiprocessing/multithreading, task scheduling, and throughput optimization. Experience with TensorRT, ONNX Runtime, Triton, Ray, Spark, or Flink is preferred.

Understanding of Embodied Intelligence

Understands the basic paradigms of embodied intelligence, VLA models, multimodal foundation models, robot learning, or imitation learning. Able to derive data collection, annotation, filtering, and evaluation strategies from model training requirements, rather than only completing isolated data processing tasks.