New Fields of investigation

HRL FOCUSES ON THE DEVELOPMENT OF INNOVATIVE TECHNOLOGIES that enable smart networks and systems to reason and adapt to changes in their environments, goals, or their own capabilities. We are developing pioneering technology in video image exploitation, image processing, data and information fusion, data mining, diagnostics and prognostics. Our focus is the development of revolutionary technology to enable smart networks and systems control that can be taught, can learn from experience to improve their performance, and can intuitively interact with their users.

Under the Defense Advanced Research Projects Agency (DARPA) Biologically Inspired Cognitive Architecture (BICA) program, HRL has developed novel, neuroscience-inspired algorithms and an architecture for learning, action, and perception that have applications in computer vision, audio processing, robot motion, and reasoning. Building on the BICA work, we are investigating human-inspired control systems that exhibit fault tolerance and robustness to previously unforeseen conditions. Utilizing perception-learning-action cycles (the same core methodology that humans use when learning how to manipulate within their environment) we have demonstrated control algorithms that are able to precisely control a 3D human-like stick robot and a human-like head-neck-eye system under previously unforeseen conditions with minimal training (something traditional control algorithms cannot do).

We are also continuing to develop and advance HRL’s video processing technology. We have combined our mature SwarmVision™ object recognition technology that can, for example, classify and keep track of stationary or moving humans and vehicles in front of stationary or moving backgrounds, with neuro-inspired attention mechanisms, demonstrating a 250% speed increase with improved detection performance. We have also developed a complete neuroinspired vision system that can learn 2D-scale, 3D-scale, and rotation-invariant representations of objects and locate these objects in a scene. Its performance is comparable to some of the best machine vision algorithms on the market. We are currently working on a system that can automatically learn and recognize complex visual behaviors. This system combines SwarmVision™ with techniques for complex relationship representation with reasoning, including fuzzy graph models, “belief” networks that update based on new evidence, Bayesian networks (“belief” networks that update probabilities based on new evidence), and ARTSTORE™ networks (self-organizing networks that can correct errors based on consequences of past actions).

NEW DEVELOPMENTS

To facilitate testing, demonstration, and transition of our technologies, HRL’s Information and Systems Sciences team is also developing test beds, test platforms, and software tools. For development and test of our video exploitation algorithms, for example, HRL has set up a video surveillance lab where we can access and manipulate imagery from 45 different camera locations. Similarly, we are porting our vision and control algorithms to embedded hardware for mobile and real-time applications. We have also developed a suite of software that enables someone unfamiliar with Bayesian networks to use our technology to develop diagnostic and prognostic solutions for their systems and applications. We are developing a Neurological Systems Modeling and ARchiTecture development framework ("NeuroSMART"). This unique software framework will enable rapid development and deployment of brain-inspired algorithms for a large variety of applications (surveillance, robotics, autonomous devices) and embedded hardware platforms.

developing technologies