Michael D. Lemmon - US grants

Direct implementation of the sensory-cognitive motor triad for intelligent robotic systems will generally place severe demands on the computing, communication, and memory resources of these systems. One solution to this problem has been to distribute the triad's functionality among various physical components of the robot. The colony-style robot ?1! and its predecessor ?2! (subsummation architecture) represent related methods for achieving this functional distribution. Colony-style robots use a collection(ie., colony) of hierarchically inhibited agents to control the robot. Existing colony-style robots suffer several disadvantages limiting their utility as autonomous system. These disadvantages involve problems in determining the robot's underlying control hierarchy and the inability of existing systems to save and use prior experience. This project uses multiagent processing paradigms called multiagent search strategies of MASS algorithms ?3! to solve the deficiencies of current colony-style systems. MASS algorithms ar inspired by recent work with competitively and cooperatively inhibited neural networks ?3!?4!. It can be shown, using statistical mechanical arguments, that competitive MASS algorithms can form minimum entropy representations of density functions. This project uses that analytical formalism to develop MASS algorithms capable of learning the components and structure of the colony-style control hierarchy. The research also explores how "cooperative" MASS can be used to realize fully discretized solution of Bellman's dynamic programming equation, and uses this capability to devise a method for implementing long term memory and path planning capabilities in colony-style robots. The project will fully develop, through simulation and analysis, the MASS algorithms needed to realize memory, learning, and planning in colony-style robots. The research will also investigate extensions of this approach to more complex systems such as flexible manufacturing systems.

Fiber and integrated optical technologies have witnessed rapid development over the last decade and are finding extensive commercial applications in optical communications, sensor signal processing, and distributed control. This project will develop an instructional senior level undergraduate laboratory providing students with hands-on experience in the design, fabrication, and testing of integrated optical devices. The laboratory will serve as a model which can be easily and relatively inexpensively emulated by other universities across the country.

This research project deals with the identification of discrete event system (DES) dynamics for synthesis of supervisory intelligent controllers. Such systems represent a complex plant's behavior by modeling it as a sequence of logical events, e.g., a computer disk control system. The supervisory controller is a DES controlling a hybrid system. Investigation of efficient methods for obtaining DES plant models are carried out in this project. Specific techniques used include the optimal designs of state space partitions defining the observed events using inductive inference procedures, with an emphasis on relatively low computational burden and reduced sample complexity.*** //

9531385 Antsaklis This is a request for block international travel funds for the 35th IEEE Conference on Decision and Control (CDC) to be held in Kobe, Japan from December 11th to 13th, 1996. The CDC is one of the world's largest and most important conferences on control theory and applications. The Kobe location will be only the third time the CDC has been held outside the United States. The requested funds will be instrumental in improving the technical success of the conference by enabling participation of sufficient numbers of key U.S. authors. ***

A supervisory hybrid system is a system generating a mixture of discrete-valued and continuous-valued signals. Such systems arise when computers supervise or control complex dynamical plants. In recent years, there has been significant interest in the use of algorithmic methods for the verification and synthesis of hybrid control systems. Algorithmic methods use algorithms whose successful execution provide necessary and sufficient conditions on the system's ability to satisfy a desired behavior. Symbolic model checking for hybrid automata represent a well-known algorithmic approach to hybrid system verification. Hybrid system model checking, unfortunately, has been shown to be undecidable for many classes of hybrid systems and this negative result has greatly limited the practical potential of algorithmic methods in analyzing and synthesizing hybrid control systems. This project proposes a paradigm shift in which we use algorithmic methods to verify the existence of 'controlled' hybrid systems satisfying desired behaviors. This approach to verification is decidable for a large class of practical systems and therefore provides a breakthrough making possible the development of algorithmic methods that provide a practical means for the verification and synthesis of hybrid control systems.

Project Abstract
Proposal #0113131
Antsaklis, Panos
U of Notre Dame

The goals of this project are to develop algorithms and prototype software for the verification and supervision of hybrid embedded control systems; also for the identification of hybrid system models. This project is developing supervisory processors to supervise and control in real time the operation of large number of control processors interacting with the outside world. The control processors interact with the physical world while the supervisory processors are responsible for monitoring and maintaining the health of the
distributed control system in a highly autonomous and fault-tolerant manner. The innovative characteristics of this project are as follows: 1) the development and application of novel discrete event supervisory methods to supervise hybrid embedded systems; 2) the development and application of novel approaches to the verification and supervision of hybrid, piece-wise linear systems; 3) the development of theory and algorithms to extend these results to a class of nonlinear hybrid systems; and, 4) the development of novel
model identification methodologies and algorithms for hybrid systems. The project's approach will improve the ability of hybrid embedded control systems to deal with high complexity, undecidability and nonlinearity.

This project evaluates the extent to which ad hoc networks of embedded systems (NESTs) may be used to implement feedback control systems. Ad hoc NESTs are inexpensive and easy to deploy, but their ad hoc nature makes it impossible to realize feedback controllers in hard real-time. NEST feedback control systems, therefore, are soft real-time systems in which feedback measurements are delayed or dropped. Maintaining specified levels of overall closed loop performance in such a soft real-time environment is extremely challenging. This project will develop methods for assuring overall closed loop performance in NEST systems. This goal will be achieved by adopting a cross-disciplinary approach that integrates
research efforts in control, real-time systems, computer systems, and communications. In particular, the project tackles this problem by 1) the dynamic re-routing of connections using a polynomial extension of minimum hop routing schemes, 2) the development of novel soft scheduling methods whose behavior can be directly related to closed loop controller performance, and 3) the dynamic re-allocation of link capacities through the use of combined channel coding and ARQ methods.

The project's broader impacts will be expressed through an undergraduate level course that integrates control theory concepts with embedded system principles. Futher project impacts will be realized by coordinating this work with an existing DARPA contract that is developing middleware technologies for NEST-type systems.

Lemmon Michael
CCR-0208537
"Performance Based Soft Real-time Scheduling in Networked Control Systems"

This project is developing novel methods for the design of high performance networked control systems. The networked control system under consideration distributes controller functionality over a collection of micro-controllers that communicate over an ad-hoc communication network. The project uses a soft variation of the (m,k)-firm guarantee model to characterize task scheduling in networked control systems. The project consists of three tasks. The first task obtains upper bounds on dropout probabilities that assure robust performance of the closed loop system. These bounds define the soft (m,k)-model, which is one major emphasis of the research. The second emphasis is on the design of implementable schedulers that enforce the soft (m,k) model arising from the control design.

The project's broader impact is achieved through the inclusion of methods and concepts developed in this project in undergraduate courses that introduce embedded system principles.

This research aims at developing novel models and methodologies for designing high performance, low cost real-time control systems. Using a networked control system as the target application, the project emphasizes an integrated treatment of control and real-time system development. The main focus of this work is on scheduling approaches for dealing with uncertainty and flexibility presented in many real-time control applications. Research activities for dealing with uncertainty includes the design of adaptive scheduling strategies to enforce a Markov-Chain based constraint and the evaluation of different schedulers. To maximally exploit flexibility existing in real-time control systems, the project investigates the effect of quantization on closed-loop system performance, and devises adaptive scheduling strategies to achieve the desired system performance while reducing resource demands.

The collaborative effort by the team of control and real-time system researchers provides unique opportunities for introducing new concepts and perspectives in tackling many challenges that are inherently interdisciplinary. The goal of the project is to advance understanding of the interplay between control system design and real-time system design.

The broader impacts of this project can be felt in two main aspects. First, as an integrated part of the project, a new course is being developed at the graduate/senior undergraduate level to expose students to both the theory and the practice of designing embedded real-time control systems. The project is expected also to have impact through a collaboration in real-time control and civil engineering, in conjunction with Notre Dame's Center for Environmental Science and Technology (CEST).

Title: Scalable Decentralized Control over Ad Hoc Sensor-Actuator Networks
Summary: This project studies the extent to which ad hoc networks of sensors and actuators
may be used to implement large-scale feedback control systems. Dense ad hoc networks suffer from
significant capacity limitations that suggest scalability may only be achieved if we restrict ourselves
to localized network traffic patterns. Moreover, such sensor-actuator networks have a large number
of clients sharing the same communication channel, so that contention between different clients will
greatly reduce the amount of data that can be streamed between sensors and actuators. Maintain-
ing specified levels of overall control system performance in such a non-deterministic environment is
extremely challenging and we feel that it can only be achieved if we adopt a cross-layer approach in
which application (i.e., control) algorithms and networking protocols are intimately aware of each
others capabilities and limitations. This project investigates fundamental relationships between
control system performance and network quality-of-service (QoS). The project will use these rela-
tionships to develop decentralized control algorithms and supporting network middleware services
that can guarantee application (control) performance over extremely large ad hoc sensor-actuator
networks. The intellectual merit of the proposed work lies in its precise characterization of the
way in which network middleware impacts control system performance. The project will demon-
strate these algorithms and services on a physical testbed in which a spatially-distributed physical
plant is controlled in a decentralized manner over an ad hoc radio (RF) network of embedded
processors.
The project's broader impacts will be felt through its impact on undergraduate education at
Notre Dame as well as through on-going collaborations with civil, environmental, and mechanical
engineering researchers both in and out of Notre Dame. In particular, the project's proposed testbed
will be tied to undergraduate courses in electrical and mechanical engineering, thereby providing
students with a novel opportunity to contribute to an on-going research program. Moreover we
are collaborating with civil/environmental engineers to develop sensor-actuator networks that can
address the combined-sewer overfow (CSO) problem faced by a number of municipalities across
the Midwest and Eastern United States. The results from this project have the potential to reduce
anticipated costs associated with solving the CSO problem. The results from this project will
therefore have a broad societal impact extending well beyond the immediate project objectives.

Networked dynamical systems are found throughout the national infrastructure.
Examples of such systems are seen in the national power grid, wastewater networks, and the national transportation system. As these networks grow in size, weak coupling across the network can blossom into system wide disturbances whose effects are felt over large geographical regions. There is, therefore, a compelling national need to devise more robust and cost-effective techniques for managing such networked systems. This project addresses this need through a self-triggered approach to decentralized control. In this approach, network subsystems use their current state information to determine the rate at which information must be exchanged to assure the networked system?s global L2 stability. These rates are cast as quality-of-service (QoS) constraints on the network?s traffic. This project uses these QoS constraints to develop soft real-time algorithms for scheduling message passing in networked control systems. The novel aspect of this work is that the resulting soft real-time control system provides guarantees on application performance that have traditionally been seen in hard real-time systems. The project is transferring the self-triggered technology to the private sector through collaborations with industry on a project that uses embedded sensor-actuator networks to control the frequency of combined-sewer-overflow (CSO) events. This CSO network is a good example of a networked cyber-physical system. To support the training of engineers qualified to manage such systems, this project is developing a set of lectures on the mathematical foundations of cyber-physical systems.

CPS: Small: Dynamically Managing the Real-time Fabric of a Wireless Sensor-Actuator Network

The objective of this research is to develop algorithms for wireless sensor-actuator networks (WSAN) that allow control applications and network servers to work together in maximizing control application performance subject to hard real-time service constraints. The approach is a model-based approach in which the WSAN is unfolded into a real-time fabric that captures the interaction between the network's cyber-processes and the application's physical-processes.

The project's approach faces a number of challenges when they are applied to wireless control systems. This project addresses these challenges by 1) using network calculus concepts to pose a network utility maximization (NUM) problem that maximizes overall application performance subject to network capacity constraints, 2) using event-triggered message passing schemes to reduce communication overhead, 3) using nonlinear analysis methods to more precisely characterize the problem's utility functions, and 4) using anytime control concepts to assure robustness over wide variations in network connectivity.

The project's impact will be broadened through interactions with industrial partner, EmNet LLC. The company will use this project's algorithms on its CSOnet system. CSOnet is a WSAN controlling combined-sewer overflows (CSO), an environmental problem faced by nearly 800 cities in the United States. The project's impact will also be broadened through educational outreach activities that develop a graduate level course on formal methods in cyber-physical systems. The project's impact will be broadened further through collaborations with colleagues working on networked control systems under the European Union's WIDE project.

The objective of this program is to develop event-triggered methods for message passing in the optimization, estimation, and control of networked dynamical systems. Prior work has demonstrated experimentally that event-triggering can greatly reduce communication usage while maintaining high levels of networked system performance. The main goal of this project is to develop formalisms that better explain the reason for these benefits and to develop a more systematic approach to designing event-triggered networked systems.


The project's intellectual merit is that event-triggering provides a solid theoretical basis for the discretization of networks of dynamical systems. This basis is based upon the simple idea that messages between subsystems should only exchanged when there is novel information relevant to the
performance of the overall system. This event-triggered approach therefore has subsystems transmit information when some internal measure of that information's novelty exceeds a time-varying and state-dependent threshold. The design of these thresholds is accomplished by enforcing stability concepts
(such as input-to-state stability or input-output stability) subject to constraints on the frequency with which information can be passed within the overall system. The transformative nature of this approach lies in its potential to provide a systematic approach to the discretization of systems in a manner that goes well beyond conventional Nyquist sampling.


The project's impact will be broadened through interactions with industrial partners EmNet LLC and Odyssian LLC. EmNet LLC is interested in using event-triggered message passing on the CSOnet system, a wireless sensor-actuator network being used to control the frequency of combined sewer
overflow (CSO) events. Odyssian LLC is interested in using event-triggered methods for the intelligent control of event-triggered microgrids. The project's impact will also be broadened through interactions
with middle school students interested in robotic systems. The project's impacts will be broadened further through interactions with European researchers.

Wireless sensor-actuator networks (WSAN) are systems consisting of numerous sensing and actuation devices that interact with the environment and coordinate their activities over a wireless communication network. This project studies "resilience" in WSANs. A resilient system is one that maintains an active awareness of surrounding threats and reacts to those threats in a manner that returns the system to operational normalcy in finite time. This project's approach to resilient WSANs rests on two fundamental trends. One trend uses machine-to-machine (M2M) communication networks that promise wireless networking with greater peak bit-rates and reliability than previously possible. The other trend comes from recent ideas that use quantization and event-triggered feedback in a unified manner to reduce bit rates required by real-time control systems. This project will evaluate and demonstrate this integrated control/communication approach to resilience on a multi-robotic testbed consisting of unmanned ground vehicles. The testbed will integrate M2M communication hardware/software with a multi-robot control architecture addressing task coordination and platform stabilization.

This project broadens its impact through organizations and programs on and around the Notre Dame campus that facilitate industrial engagement and technology transfer. The project will engage undergraduate and graduate students to support the project's testbed and algorithm development. The project will augment and re-organize Notre Dame's Cyber-Physical System (CPS) curriculum by integrating the results of this project into courses.