Model Algorithm Engineer — World Model Direction

Develop world-model algorithms and training data systems for embodied intelligence.

What You Will Do

World Model Algorithm R&D and Training

Develop, train, and evaluate world-model-related algorithms for embodied intelligence scenarios, including environment representation, state prediction, video/trajectory prediction, multimodal modeling, and action-conditioned modeling.

Data Processing and Cleaning

Handle cleaning, filtering, annotation-rule design, and quality evaluation for multi-source robot data. Identify and resolve issues such as data noise, distribution bias, abnormal trajectories, and invalid samples.

Training Data Construction and Iteration

Based on model training results, participate in the design and optimization of data pipelines. Support data format conversion, sample splitting, data alignment, feature extraction, and training/validation set construction, continuously improving data quality and model performance.

Collaboration with Robot Models and Systems

Follow and participate in the development and deployment of robot end-to-end manipulation models, vision-language-action models, and vision-language-navigation models. Combine real robot data and task feedback to promote the application of algorithms in robot manipulation, planning, control, and policy learning.

Low-Level Algorithm and Engineering Implementation

Participate in the development and maintenance of model training code, data processing scripts, evaluation tools, experiment management workflows, and training pipelines. Strong coding ability is required, with the ability to efficiently complete experiment development, data processing, and troubleshooting using AI programming tools.

Large-Scale Training and System Optimization

Participate in large-scale multimodal model training, distributed training, model parallelism, data parallelism, training performance optimization, and massive data processing according to project needs.

What We Expect

Academic Background

Background in computer vision, artificial intelligence, robotics, computer science, automation, mathematics, electronic information, or related fields. Bachelor’s degree or above. Graduates from domestic Top 5 universities or renowned overseas universities are preferred/required.

Programming Ability

Proficient in Python and familiar with deep learning frameworks such as PyTorch and JAX. Strong programming and engineering skills.

Low-Level Practical Experience

Experience with model training frameworks, data pipelines, evaluation systems, simulation environments, robot data collection, or real-system deployment is preferred.

Data Processing Experience

Practical experience in data cleaning, data processing, dataset construction, training data analysis, or data pipeline development. Able to understand the impact of data quality on model performance and complete full data-processing and training loops.

Model Understanding

Deep understanding of world models, generative models, or sequence modeling. Familiar with one or more methods such as Diffusion, GAN, Flow, Transformer, or VAE. Large-scale training experience is preferred.