Research is performed to establish the role of innate theories in cognitive robotics. Current architectures have a major weakness: the semantic levels, although perhaps patterned after biological systems, are selected in an ad hoc way. This project develops domain theories for the following levels of cognitive function:
? self-knowledge: ? dynamical vehicle control: rigid body motion is fundamental to dealing with the physical world, and Lie algebra is used as the key domain theory.
? particular constraints ? rules of the game (e.g., traffic laws): rule-based systems provide the underlying methodology for understanding rules.
? user interaction ? agent behavior: graph eigenvector clustering is used to segment basic behaviors of other agents, including people.
? operational context ? active embodiment computation and communication control: high-level policies on latency, throughput, priority, and parameter selection provide a domain theory.
Specifically, the project studies how symmetry theory can be exploited in a novel perception-action architecture. Symmetry plays a deep role in our understanding of the world, in that it addresses issues of invariance. The determination of operators which leave certain aspects of state invariant makes it possible to identify objects or maintain specific constraints while performing other actions. Symmetry operators in signal analysis, concept formation, learning and platform control are studied. The potential impact of the work is a more cohesive, expressive, adaptable and effective cognitive robotics framework. Results will be disseminated at robotics and AI conferences as well as at international workshops (e.g., Schloss Dagstuhl). Potential applications include: cognitive vehicles, sensor networks, buildings, interactive toys, etc.
Innate Theories in Cognitive Robots Final Report for NSF Award ID 1021038 Thomas C. Henderson, PI 7 May 2012 1. Project Outcomes and Findings This EArly concept Grant for Exploratory Research allowed investigation of a new approach to cognitive robotics. We studied the state of the art in embodied, networked cognitive systems and developed new targeted domain theories which are proposed as innate to the cognitive system. A cognitive agent optimizes its behavior to achieve an objective efficiently by finding models (symmetries) that resolve hidden state information and that help it to predict the future under a variety of real-world situations. These processes involve monitoring, exploration, logic, and communication with other agents. As a specific application, we focused on mobile robots as the first step to demonstrate the feasibility of the approach. We made the following key technical advances: (1) developed domain theories of symmetry, (2) tested these theories on actual sensor data, and (3) measured the performance to demonstrate effectiveness. The exploitation of symmetry analysis may be widely applicable to complex cognitive systems including those for autonomous wheel chairs, structural health monitoring, and robot surgery, for example. 2. Publications "Symmetry: A Basis for Sensorimotor Reconstruction," Thomas C. Henderson, Hongchang Peng, Christopher Sikorski, Nikhil Deshpande, and Edward Grant, University of Utah Technical Report, 11-011, May, 2011.
