Wiesner Room, 2nd Floor, MIT Media Lab
From an early age, humans are able to use intentions and beliefs to understand the behavior of others. While these qualities are not directly observable, they can be inferred from observable behavior and carry great predictive power. This thesis presents a robotic architecture designed to work with the concepts of intentions and beliefs to more efficiently and naturally interact with humans, both by modeling the intentions and beliefs of the human, and by being aware of how the human is also doing the same about other agents including the robot itself. Through its realization in an embodied, real-time, robotic system, this architecture focuses on the connections between the hidden states and the observable world, including visual perspective and action performance. By adding an ability to carry these processes into the short-term future, the robot gains the capacity to interact directly with human mental states, making and carrying out plans that address their formation or alteration.
This thesis presents a series of architectural mechanisms which allow the robot to pass the false belief test, infer goals from failed actions, and take action to manipulate mental states. These skills are demonstrated using the robots Leo and Nexi.
Host/Chair: Cynthia Breazeal
Deb Roy, Bruce Blumberg