Our core research areas include:

• Trusted Generative AI for Human Values: Pioneering high-fidelity generation across text-to-image, text-to-video, and text-to-3D modalities with advanced personalization techniques, while ensuring fairness and eliminating bias.
• Perception & Reasoning for Safety-Critical AI: Developing safe AI systems that perform under real-world challenges such as low light, adverse weather, autonomous driving, and other safety-critical applications.
• Physical Grounding & Spatial Intelligence: Bridging MLLMs with the physical world by teaching 3D spatial reasoning and physical laws, and advancing fine-grained spatial understanding for precise interaction between AI and its environment.
• Efficient & Scalable AI: Creating resource-efficient AI through lightweight training, model compression and quantization, and dataset distillation.