From collaboration with the University of Waterloo in human experiments, to understanding of the brain's perceptual and decision making apparatus, to realization of computational models in robots.
Neuromorphic Computing is an event based system which mimics spike based communication between neurons in the brain.
We pioneered using mutually connected spiking neurons to control real robots. We also collaborated in building spiking aVLSI circuits into a "CPG-Chip"
We have collaborated with neuroscientist mutally entrain our chip with the spinal cords of animals.
We have built circuits which can drive under-actuated, robots that feature biarticular 'muscles'
Learning is a key feature of intelligence.
We were the first to adapt a neural network to control a real robot in real-time 30 years ago.
We also explored how to use a collection of simple learning modules, that when sequenced could program a robot to walk. We used an abstraction of neural modules called "ring-rules" which sit between limit-cycle oscillators and decision trees.
We developed neural perceptual predictors for locomotion to extract novelty from a visual stream. This is patented work.
We co-invented the "virtual structures" method of formation flying. Given a any number of robots, maintain a formation using he biological notion of heterarchical control- the antithesis of hierarchical control fist highlighted by Avis Cohen.
Applications: Large baseline space telescopes, drones in formation, manipulation of payloads using multiple drones
Behavior based control system intertwines planning, reactivity and control.
Decades ago, I directed my first lab soon after undergrad where I designed and built a robotic snake as part of an Internation Research and Development effort at Hughes.
Latter at USC I started building autonomous helicopters. One helicopter was all electric. I got the idea for building it from Chris Atkison at CMU. At that time electric rotor flight was in its infancy. I also experimented with simulations of 4 blade machines, what are the common drone configuration today. I realized they had much simpler dynamics than helicopters. One helicopter design went on to win an international competition and the project went to NASA/JPL under my colleague Dr. James Montgomery.
At NASA/JPL I built a distributed embedded DSP control system for a 50 Degree of Freedom hand-controller/exoskeleton system. I got to build a bare metal system starting with a boot loader all the way up to optical network packet-based communication with a stable 1,000 Hz bandwidth. I got a Nasa New Technology award.
My students and I have designed and built a number of walking machines exploring neural control and biarticular muscle architectures.
