Biographical Sketch

Caltech Research

Curriculum Vitae (pdf)

Publication List (pdf)

Online Publications

PhD Alumni

Contact Information

Sailing!

home

Rod Goodman B.Sc., Ph.D., C.Eng., FIET, FIEEE.

Caltech Research

My research at Caltech was generally in the area of intelligent information processing systems. The goal was to design autonomous adaptive systems that learn and exhibit intelligent decision-making behavior, the emphasis being on systems. Only in a complete system do all the pieces come together: hardware, software, algorithms, and architecture. Each must work in perfect unison to obtain intelligent systems level behavior. The systems that I worked on at Caltech are in the areas of Robotics and Automation , Neuromorphic VLSI Processing , Information Processing Systems , and Cognitive Systems .

My work in Robotics at Caltech focused on Autonomous Collective Robotics . Here we take inspiration from biology in the way that simple low level sub-system units (such as ants) can lead to complex systems level behavior (such as the functioning of a complete nest).  We ask the question: how can we apply these principles to engineering systems? In particular, how do we design low-level behaviors for simple small robots that interact with each other and the environment to produce complex systems level behavior?    The results of this work are not only applicable to robotics. For example, we have applied these principles to routing in communications networks, and to the control of automotive traffic and satellites. At Caltech, with the help of Alcherio Martinoli (EPFL), Owen Holland (Sussex University), and Alan Winfield (University of the West of England), I built up a research group that focused on autonomous collective robotics. In addition to setting up a robot laboratory with a large number of robots, in 2000 we developed and taught the first ever course on Swarm Intelligence.

My second “biologically inspired” area of interest at Caltech was that of Neuromorphic VLSI Processing sub-systems. These are systems that emulate the neural processing that takes place in biological systems, such as those that give us our senses of vision, sound and speech, touch, and chemical sensing. The performance of these biological systems is truly impressive and motivated our research into the application of these principles to engineering systems for sensing and actuation.  The Neuromorphic approach was inspired by Carver Mead and John Hopfield and led to the formation of Caltech's Computation and Neural Systems (CNS) degree program. Within CNS my group developed the Caltech Electronic Nose, and the Caltech Silicon Active Skin – a project that combined sensing, processing, and actuation on an integrated CMOS/MEMS system chip. In addition, my Neuromorphic VLSI Processing group researched and demonstrated VLSI systems for on-chip neural network learning, high capacity neural associative memories which can perform real-time vector quantization on images, as well as chips for video contrast enhancement, color processing, and stereopsis.

My third research area at Caltech was in the area of Information Processing Systems . This work is more “algorithmically based” and built on my past work in Neural Networks, Machine Learning, and Information Theory and Coding. It focused on building system applications such as: neurocontrollers, texture recognition, image compression, face recognition, handwritten document analysis, keyword spotting in historical cursive documents, data visualization, information filtering, and smart searching on the Web. In particular, my information processing research has resulted in an information theoretic approach to modeling decision systems which extends from the automated capture of knowledge in the form of rules, to the implementation of Bayesian probabilistic inference systems on multi-layered parallel neural-network architectures. Thus, the machine learns automatically from experience, executes its knowledge in true real-time and its decisions are understandable to humans because the knowledge is in the form of human-readable “rules”. This approach was also applied to modeling and controlling telephone traffic, fault finding, stock market prediction, building neural networks that can discover state machines and grammars, and networks that can analyze music and then create new music. We also embedded these techniques into software agents that could autonomously search and transact business on the web. Neural network algorithms for control and diagnostic applications were also investigated. Neural networks are capable of learning complex non-linear dynamics, which are difficult to model mathematically. In particular, we developed hybrid neurocontrol techniques that combine linear and non-linear control theory, reinforcement learning and case based reasoning to produce more robust controllers than are achievable with one technique alone.

My fourth research area at Caltech was that of Cognitive Systems. I am interested in discovering the “reason” for consciousness in higher mammals, and why biology has evolved this system for the control of complex organisms.  The goal is to then use these principles as the guiding architecture for complex cognitive robots and other autonomous machines. In collaboration with Christof Koch, David Chalmers, and Owen Holland , I co-organized the first ever conference on Machine Consciousness which was held at the Banbury Center of Cold Spring Harbor Laboratory in 2001.

top