Jeffrey L. Krichmar, Ph.D. - US grants

EMT/BSSE: A Controller for Autonomous Systems
Based on Principles of Vertebrate Neuromodulation
Robots and autonomous systems require some level of supervision and tuning of
parameters to fit a particular domain. However, biological organisms have the ability to
respond quickly and appropriately in an ever-changing world. Because this adaptability
is so critical for survival, all vertebrates have sub-cortical structures, which comprise the
neuromodulatory systems. These systems regulate fundamental behaviors and set the
organism?s internal and behavioral states. This research involves designing a controller
for autonomous systems that is modeled after the vertebrate neuromodulatory system.
This neurally inspired model enables robots to approach the behavioral complexity and
flexibility associated with higher order animals, and would be a major improvement in the
design of autonomous systems.
Neuromodulators in the nervous system signal environmental changes to the nervous
system that alter neuronal responses in such a way that the organism can respond
quickly and accurately to these changes. There are separate neuromodulators that
respond to threats, reward anticipation, novelty, and attentional effort. However, each of
these neuromodulatory systems have a similar effect, that is, to cause an organism to be
decisive when environmental conditions call for such actions, and allow an organism to
be more exploratory when there are no pressing events. A design strategy, based on
principles of the vertebrate neuromodulatory system, is used to control the behavior of
autonomous robot systems. This research shows that such a system responds
appropriately and adapts to environmental changes without human intervention.
Moreover, this research shows that a controller, which is based on neuromodulation, can
be extended to any system in which an agent is situated in a dynamic, unconstrained
environment.

Humans, rats and other vertebrates, relying on their advanced nervous systems, are far superior at dealing with the uncertainties of the world than are artificial systems. Thus, a machine, whose behavior is guided by a neurobiologically inspired system, might demonstrate the flexible, autonomous behavior normally attributed to biological organisms. Biological organisms have the ability to respond quickly to an ever-changing world. Because this adaptability is so critical for survival, all vertebrates have sub-cortical structures, which comprise the neuromodulatory systems, to handle uncertainty and change in the environment. Attention, which is influenced by neuromodulation, plays a significant role in animal's ability to respond to such changes. Different neuromodulatory systems are thought to play important and distinct roles in attention. A collaborative approach, which compares rodent experiments with robots having simulated nervous systems, will examine these attentional systems. These experiments will lead to a better understanding of how animals cope with uncertainty in the environment, and will lead to the design of a robot capable of flexible and complex behavior. This work has the potential of being paradigm-shifting technology that could find its way in many practical applications.

In an interdisciplinary approach, a robotic system, whose design is based on the vertebrate neuromodulatory system and its effect on attention, will be constructed and tested under similar experimental conditions to the rat, and then in a more practical application. This approach, which combines computational modeling and robotics with rodent behavioral and electrophysiological experiments, will lead to a better understanding of how areas of the brain allocate attentional resources and cause the organism to respond rapidly to essential events and objects. Two of these neuromodulatory systems, the cholinergic and noradrenergic, are thought to play important and distinct roles in attention. Expected uncertainty, the known degree of unreliability of predictive relationships in the environment, drives activity within the cholinergic system. Unexpected uncertainty, large changes in the environment that violate prior expectations, drives activity within the noradrenergic system. These systems modulate activity in brain areas to properly allocate the attention to stimuli in the environment necessary for adequate learning to occur and fluid behavior to be maintained. This knowledge will be used to construct a robust, intelligent robotic system whose capability to adapt to change, and behave effectively in a noisy, complex environment will rival that of a biological system.

A framework will be developed to help scientists and engineers create brain-inspired, brain-sized networks that can carry out practical applications. Large-scale spiking neural networks, which follow the brain's architecture and activity, have been used to successfully model phenomena such as learning and memory, vision, auditory processing, neural oscillations, and many other important aspects of neural function. Additionally, spiking neural networks are particularly well suited to run on neuromorphic hardware, state of the art computers that emulate the brain?s structure and dynamics. These neuromorphic systems depend on the binary nature of spikes to lower communication bandwidth and energy consumption. Although significant progress has been made towards the specification and simulation of large-scale spiking neural networks on a variety of hardware platforms, many challenges remain before these neurobiologically inspired algorithms can be used in practical applications. While biology does provide increasingly abundant empirical data that can constrain these systems, many parameter values must be chosen manually by the designer to achieve appropriate neuronal dynamics, a task that is extremely tedious and often error-prone. To meet this challenge, an automated parametertuning framework will be developed that is capable of quickly and efficiently tuning large-scale spiking neural networks. The framework will leverage recent progress in evolutionary algorithms and optimization techniques for off-the-shelf graphics processing units (GPUs). The parameter search will be guided by the idea in neuroscience that biological networks adapt their responses to increase the amount of transmitted information, reduce redundancies, and span the stimulus space. This notion of efficient coding will guide the tuning process of the artificial spiking neural networks. Computer scientists and engineers will be able to use the resulting automated parameter-tuning framework to create brain inspired applications, such as vision and memory systems, on neuromorphic hardware. Moreover, the resulting framework will allow neuroscientists to more readily create models that better describe their empirical data and generate new quantitative hypotheses that can be tested in the laboratory.

This project develops efficient machine vision algorithms inspired by the architecture and energetic efficiency of the primate visual system for motion processing. Navigating through a rich cluttered natural environment, while both the observer and the objects in the scene are moving, is a difficult problem in machine vision, particularly for real-time processing under power constraints. However, humans and other animals perform these tasks with ease. The nervous system is under tight metabolic constraints and this leads to incredibly efficient representations of important environmental features, such as the observer's heading, the depth of objects, and the motion of objects. In addition, these efficient machine vision algorithms can be applied to robotics, the IoT, and edge processing. The algorithms can be applied to a wide range of applications, including augmented reality, assistive robotics, autonomous vehicles, and the Internet of Things (IoT) Thus, they could have a transformative economic and societal impact by creating applications that can operate autonomously over long periods in remote locations.

Inspired by ability of the nervous system to efficiently encode and appropriately respond to the visual features that make up a dynamic scene, the algorithm uses sparse predictive coding techniques to process data streams from cameras. Because the algorithms can be realized in spiking neural networks, where the artificial neurons only send signals when an event occurs, they can run efficiently on low powered neuromorphic systems; computers that support such representations. By employing an architecture inspired by the brain, where op-down signals from the frontal cortex and parietal cortex predict where objects will be in the future, the system will have better object tracking and overcome difficulties when objects become hidden from view. These representations are sparse and reduced, leading to energy efficient processing, less computation, and thus low power consumption. In summary, the machine vision algorithms: (1) increase our understanding of how the brain encodes behaviorally relevant signals in the world, (2) lead to computationally efficient handling of large data streams, and (3) realize power efficient processing for a wide range of embedded applications.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.