Description
The role will focus on the following research directions:
Embodied Intelligence Foundation Models
Construct ultra-large-scale, high-quality pre-training datasets for embodied intelligence. Research and develop next-generation multimodal (visual, linguistic, force, tactile, etc.) fused foundation models for embodied intelligence. Conduct model pre-training and fine-tuning to unlock emergent capabilities, enabling deep understanding of the physical world and general reasoning competence.
Ultra-High-DOF Dexterous Locomotion and Manipulation
Develop full-body cooperative control algorithms based on advanced AI paradigms including deep reinforcement learning and imitation learning. Realize dynamic balance, agile locomotion and fine manipulation of robots (e.g., humanoid robots, dexterous hands) in complex and unstructured environments. Tackle manipulation tasks involving rich contact forces, and achieve seamless execution of dexterous manipulation during robot motion.
Human-Robot Cross-Embodiment Skill Transfer
Develop innovative cross-embodiment imitation learning and correspondence learning algorithms. Enable robots to comprehend and replicate skill demonstrations from different embodiments via visual teaching, linguistic instructions and other modalities. Research skill and knowledge transfer across agents with diverse morphologies and dynamic structures (e.g., humans, humanoid robots, wheeled robotic manipulators, etc.).
Requirements
- PhD candidates or postdoctoral researchers graduating in September 2025 or later, or individuals with equivalent research depth.
- Publications in top-tier robotics conferences or journals (e.g., Science Robotics, IJRR, TRO, RSS, CVPR, ICLR, etc.).
- Participation and award-winning experience in prestigious robotics competitions (e.g., RoboMaster, RoboCup, etc.).
- Other high-value research achievements, project deliverables or publications.
