Computer Vision Lab (CVLAB)

Physical AI bridges artificial intelligence with the physical world through robotics and embodied systems. We develop agents that can perceive, reason, and act within real-world environments. Our research covers motion planning, manipulation, and human–robot interaction. By combining learning with physical embodiment, AI systems gain adaptive and robust capabilities. This work aims to create intelligent machines that seamlessly integrate into human society. Agentic AI explores autonomous systems that can reason, plan, and act toward achieving complex goals. Our research focuses on building AI that moves beyond passive prediction to proactive decision-making. We study goal-directed planning, adaptive reasoning, and interaction with dynamic environments. By integrating autonomy and continuous learning, agentic AI can adapt to real-world challenges. This work enables next-generation intelligent agents for science, robotics, and digital platforms. LLM-Driven Multimodal Models integrate large language models with vision, audio, and other modalities. Our research explores how LLMs can serve as a unified reasoning engine across diverse inputs. We develop models that align language with perception for richer understanding and generation. This includes tasks such as multimodal reasoning, grounding, and cross-modal interaction. Our goal is to enable versatile AI agents that seamlessly bridge language and the real world. AI Safety ensures that artificial intelligence systems are reliable, ethical, and aligned with human values. Our research investigates risks such as bias, misuse, and unintended consequences of AI. We study robust training, interpretability, and alignment methods for trustworthy AI development. By building safeguards and evaluation frameworks, we aim to prevent ha