About the Project
Embodied AI is rapidly transforming the future of robotics by enabling robots to perceive, reason, learn, and interact with the physical world. Recent advances in foundation models, Vision-Language Models (VLMs), and Vision-Language-Action (VLA) models have significantly enhanced robotic capabilities in perception and decision-making. However, ensuring safety, reliability, and trustworthiness remains one of the most critical challenges for deploying robots alongside humans in real-world…
environments.
This PhD project aims to develop next-generation Safe Embodied AI systems for human-robot collaboration and humanoid robotics. The research will investigate how multimodal foundation models can be integrated with human-aware planning, robot learning, and provably safe control to enable robots to safely and efficiently collaborate with humans while performing complex tasks in dynamic and uncertain environments.
The project will explore novel approaches that combine foundation models, reinforcement learning, optimization, game theory, and safety-critical control to improve robot perception, decision-making, adaptability, and safety. Particular emphasis will be placed on enabling robots to understand human intentions, anticipate human behavior, learn from interactions, and operate safely in close proximity to people.
Potential research topics include:
- Vision-Language-Action models for robotics
- Human intention understanding and prediction
- Human-aware motion planning and decision making
- Safe reinforcement learning and continual learning
- Control Barrier Functions and safety-critical robot control
- Human-robot collaboration and shared autonomy
- Multi-agent and multi-robot embodied intelligence
- Safe and trustworthy humanoid robotics
- Long-horizon embodied task planning and execution
