Cognitive Robotics

We believe that to create truly intelligent autonomous robots that will collaborate with a human, the robots must use similar representations and similar reasoning mechanisms to a human. In this work we use a computational cognitive modeling tool, ACT-R, to develop reasoning and representations based on human models. This has been applied to tasks such as hiding and seeking, and to a very critical ability for collaboration -- perspective taking.

Human-Robot Interaction

Development of principles, and techniques for better interaction between humans and dynamically autonomous robots (including autonomous vehicles). Techniques under investigation include: multi-modal interfaces that combine speech, gestures, PDAs and other modes of interaction; the use of computational cognitive models of human behavior as reasoning mechanisms for dynamically autonomous robots; and development of dynamic autonomy and mixed-initiative interaction. We are investigating these interactions in the context of teams of humans and multi-robot systems.

Continuous and Embedded Learning

Intelligent, dynamic autonomy will require systems that can adapt to changes in the environment and to changes in their own capabilities, as well as to changes within the team in which they are a member. We are investigating several hybrid machine learning techniques, and various representations, as well as designing a system that allows for safely learning and adapting while in the real world.

Adaptive Testing

In this work, we use machine learning techniques to automate robustness testing of complex systems such as intelligent controllers for autonomous vehicles. Our learning system, which is based on a combination of evolutionary algorithms, reinforcement learning, and lamarckian learning, explores the space of fault scenarios that cause maximum failure of the vehicle with minimum faults. The technique has been used to test intelligent controllers for simulated unmanned air vehicles and on autonomous underwater vehicles.