Research Thrust 2: Embodied Multi-Agent AI
This thrust addresses how large populations of embodied agents learn, coordinate, and act under uncertainty, communication constraints, and safety requirements. Building on Phase I advances in distributed optimization, multi-robot systems, and networked learning, it uses multi-modal world models and decentralized optimization for federated learning, sensing, and safe exploration.
TILOS Publications on Google Scholar (opens in a new tab)
Theme E1Embodied World Models
Embodied agents that construct and refine world models through physical interaction, fusing visual, proprioceptive, tactile, and other sensing streams.
Research directions
- World models for physically embedded perception, control, and prediction
- Multi-modal probabilistic fusion under real-time constraints
Faculty
Henrik I. Christensen, Nikolay Atanasov, Tara Javidi, Vijay Kumar, Camillo J. Taylor, Xiaolong Wang
Project highlights for this theme are coming soon.
Theme E2Multi-agent Optimization
Distributed optimization and graph-based coordination that let agents with partial observations and limited communication converge on globally coherent behavior.
Research directions
- Physically embedded federated sensing, learning, inference, and control
- Decentralized active sensing, exploration, and learning
Faculty
Vijay Kumar, Nikolay Atanasov, Henrik I. Christensen, Sicun Gao, Tara Javidi, Alejandro Ribeiro
Project highlights for this theme are coming soon.
Theme E3Safety and Trust for Multi-agent Systems
Integrating formal safety guarantees, learned behavior, and human oversight into a unified framework for embodied multi-agent systems.
Research directions
- Safe and trustworthy reasoning for multi-agent systems
- Human–robot collaboration
Faculty
Tara Javidi, Sicun Gao, Farinaz Koushanfar, Vijay Kumar, Alejandro Ribeiro, Xiaolong Wang
Project highlights for this theme are coming soon.