Large Model Algorithm Engineer - Embodied AI

Skills

About the Role

You will work on the brain for embodied intelligence, developing multimodal models that understand spatial relations and state changes from images, video and language, and decompose goals into executable steps. You will explore long-horizon goal understanding, task decomposition, planning and error correction, take part in multimodal data construction and model training using supervised fine-tuning and reinforcement learning, and build reproducible evaluations covering spatial reasoning, long-horizon completion, failure recovery and cross-scenario generalization.

Requirements

• Solid machine learning and deep learning foundation; familiar with Python and PyTorch; able to independently reproduce papers, train models, analyze experiments and debug
• In-depth research or engineering practice in at least one of embodied AI, robot learning, VLA, world models, vision-language navigation or multimodal spatial reasoning
• Experience with training and evaluating large or multimodal models: pre-training, post-training, reinforcement learning, imitation learning or multimodal data construction
• Good research judgment and engineering ability, turning open problems into verifiable hypotheses

Responsibilities

• Develop multimodal large models for embodied tasks, improving spatial understanding, temporal reasoning, state tracking and environment change prediction
• Explore goal understanding, task decomposition, action planning and feedback correction in long-horizon tasks
• Participate in multimodal data construction and model training, designing data recipes and training tasks, exploring supervised fine-tuning and reinforcement learning
• Build reproducible embodied capability evaluations and drive iteration of data, training strategies and models