Note: This content is accessible to all versions of every browser. However, this browser does not seem to support current Web standards, preventing the display of our site's design details.


Model-Based Autonomous Control

Many modern artifacts, such as automotive systems, airplanes, mobile robotic devices and space probes, exhibit complex patterns of behavior in order to satisfy the high demand on performance, durability and autonomy. Key for the artifact’s operation is a sophisticated hybrid control system that orchestrates the many components of the artifact through closed loop, system-wide interaction. Most model-based approaches in control use an artifact model to synthesize state observers and control laws that are executed on-line. Thus, the control objective is stored implicitly, which makes it difficult for the control system to adapt to changing operational conditions or (possibly un-foreseen) system faults. Our approach is different in that we use a hybrid stochastic model of the artifact and a declarative description of the control goal to ‘program’ the control system and provide computational methods that efficiently deduce on-line state estimators and hybrid control laws. To achieve this goal we work on a tool-set for hybrid control that builds upon methods from planning and control in AI, model-based reasoning and on-line control optimization methods such as model predictive control. More specifically, we work on methods for: Configuration Management: A complex artifact with built-in redundancy allows several configurations to achieve it’s operational goals. For example, a mobile robot with redundant actuation provides several configurations of its drives to achieve mobility. Configuration management deduces a suitable configuration on the basis of the artifact’s hybrid model whilst respecting operational constraints, for example, due to faults that are identified through hybrid diagnosis. Hybrid State Estimation and Diagnosis: A key functionality of autonomous control is to track the artifact’s behavior as it dynamically moves through a possibly large set of operational and fault modes. Computational complexity of this task prevents one from considering all possible evolutions so that sub-optimal methods that focus onto a set of most likely modes are needed. Our approach combines multi-model filtering methods with advanced search and reasoning methods from the toolkit of AI and provides means for on-line system-analysis, system-decomposition and filter deduction that handles systems with a potentially large number of operational modes and failure. Hybrid Control Law Deduction: Planning for a complex hybrid control strategy that actuates an artifact according to a control objective can overwhelm traditional optimization methods of control theory because of a potentially large number of operational mode sequences. We utilize qualitative reasoning methods to pre-select among suitable control sequences and thus focus continuous control methods to feasible solutions of the control problem. In my talk, I want to detail our overall approach for model based autonomous control through demonstrating solutions for the tasks outlined above and show their interplay within the overall control scheme.

Type of Seminar:
Public Seminar
Prof. Dr. Michael Hofbaur
Technische Universität Graz, Austria
Mar 10, 2006   16:15

ETH Zentrum, Physikstr. 3, Building ETL, Room K25
Contact Person:

Prof. M.Morari
No downloadable files available.
Biographical Sketch:
Michael Hofbaur is associate professor for automation at the Institute of Automation and Control, Graz University of Technology, Austria. From 2000 to 2001, he was with the Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, USA. Dr. Hofbaur's research objective is to provide methods for autonomous automation of complex systems that are on the interface between control theory, computer science and artificial intelligence. The goal is to build autonomous artifacts that reason quickly, extensively and accurately about the world and react to novel or unforeseen situations. For this purpose, he conducts research in the fields of qualitative and model-based reasoning, hybrid system theory, and model-predictive control.