Multi-Strategy Multi-Functional Intelligent Systems

Multi-strategy reasoning refers to the use of different reasoning methods, for example, the case-based and model-based methods of reasoning. Multi-strategy reasoning is a common theme in many of our projects including KRITIK, IDEAL, AUTOGNOSTIC, and especially ROUTER. Multi-functional knowledge systems are capable of addressing different reasoning tasks such as adaptive device design and natural language understanding.

Multi-Strategy Systems - Router

Our work on multi-strategy reasoning has led to a general architecture for dynamic and flexible strategy selection and integration in multi-strategy knowledge systems. ROUTER, a qualitative navigation planning system, instantiates this architecture. The system integrates the methods of case-based plan reuse and model-based search for planning navigation paths in physical spaces. Experiments with ROUTER's architecture results in robust, flexible and opportunistic reasoning. Current work focuses on incorporating additional computational constraints on strategy selection.

Multi-Functional Systems - KA

The KA project explores the design principles for building multi-functional knowledge systems. KA both interprets design requirements stated in English and solves the interpreted design problem. It uses the same SBF ontology for low-level language processing, high-level conceptual understanding, and design problem solving. In addition, it uses the same knowledge (design cases and device models) and methods (case-based and model-based reasoning) for high-level language understanding and design problem solving. Current work focuses on the acquisition of device models from English texts.

Large-scale Knowledge Systems - HIPED

The key problem in designing large-scale knowledge systems is heterogeneity, where the heterogeneity may lie in the types of knowledge, forms of representation, reasoning strategies, and/or reasoning tasks. The HIPED, ROUTER, and KA projects explore different kinds of heterogeneity. The ROUTER project investigates the integration of multiple reasoning strategies while the KA project investigates the use of common knowledge, representations, and strategies to support multiple reasoning tasks such as natural language understanding and adaptive design. While the ROUTER system has led to a general architecture for flexible strategy selection and integration, KA has led to a common ontology for some kinds of language-understanding and adaptive-design tasks. HIPED investigates the issues of accessing and integrating relevant knowledge from heterogeneous information sources. The current focus of this work is on the incremental compilation of retrieval knowledge and strategies into meta-models and meta-cases for use in subsequent knowledge queries.

Interaction between Deliberative Planning and Reactive Control - RAURA

An interesting problem in robotics concerns the interaction between deliberative reasoning and reactive control. The RAURA and REFLECS projects explore this interaction. RAURA is an instantiation of ROUTER in the AuRA architecture (AuRA is a hybrid architecture for autonomous mobile robots capable of both deliberative planning and reactive control developed by the Robotics group). RAURA itself is embodied in a physical robot. Experiments with RAURA show that a qualitative symbolic navigation planner, such as ROUTER, coupled with numerical reactive control is sufficient for navigating some kinds of engineered (but real) environments. Current work focuses on further integration of qualitative symbolic approaches to deliberative planning and numerical methods for reactive control.

Interaction between Deliberative Learning and Reactive Control - REFLECS

REFLECS is a reflective system capable of learning strategies for reactive control by diagnosing and repairing failed reactive strategies. REFLECS instantiates the Autognostic system in the AuRA architecture. Experiments with REFLECS show that for some kinds of failure of reactive strategies (e.g., when two reactive strategies are in conflict), a model-based deliberative learning system, such as AUTOGNOSTIC, can help diagnose and repair the failed reactive strategies. Current work focuses on further integration of deliberative learning and reactive control.

Some Related Papers

Some Experimental Results in Multistrategy Navigation Planning, Ashok K. Goel, Khaled S. Ali, and Eleni Stroulia. GIT-CC-95-51.

Multi-Strategy Reasoning Papers



For links to a complete list of papers and some ftp'able versions of the above papers go to: Complete Listing