Emergent Behaviors in Multi-Agent Human-Robot Teams
Main content start
Principal Investigators:
Stefano Ermon and Dorsa Sadigh
TRI Liaison:
Project Summary
Develop data-driven models of the latent structure behind human decision making in the context of multi-agent mixed-autonomy interactions (e.g. traffic intersections, shared roads, human-robot teaming, and mutual adaptation).
Models of people’s intentions and interactions will improve safety and reliability and may open new avenues for human-robot interactions and teaming. Real world settings include unprotected left turns, merging, etc.
Research Goals
Develop variable generative models to:
- Generate multi-agent behaviors indistinguishable from training data.
- Perform multi-agent intent inference in an unsupervised way.
- Learn teaming behaviors and adaptations.
Models will be evaluated in simulation and using real-world data.