Center for Autonomy Computing (CAC)

Autonomy functions focus on perception and control of autonomous systems, signal processing, and robotic swarm behavior.

Perception

The information & Systems Sciences Lab has extensive experience in intelligence, surveillance, and reconnaissance technologies, including bio-inspired attention, neuromorphic object recognition, 3D recognition using LIDAR, tracking, and sensor fusion. For applications like autonomous driving and flight, we are developing more accurate object and situation recognition systems that are robust to errors in individual sensors.

Selected Publications:

• D. Khosla, Y. Chen, and K. Kim. A neuromorphic system for video object recognition. Frontiers of Computational Neuroscience, November 28, 2014

Signal Processing

HRL’s Cognitive Signal Processing technology combines innovations from machine learning, dynamical system theory, and real-time optimization to perform rapid short-term prediction, anomaly detection, and de-noising of wideband signals in less than 10 nanoseconds. These algorithms are highly scalable (both in bandwidth and complexity) and have been successfully demonstrated to reduce noise by 20-40dB on a wide variety of challenging synthetic and experimental RF data sets.

This material is based upon work supported by the Office of Naval Research, under contract number N00014-12-C-0027. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Office of Naval Research.

Swarm Behavior

Accurate prediction of swarm behavior is a crucial step in counter-swarm tactical adjustment. To this end, HRL’s new innovations in swarm technologies has been extended to make predictions of multi-agent adversarial movements. We have introduced different techniques and evaluate them with real-world datasets from team sport events, predicting, e.g., adversary basketball player positions with less than 87cm mean square error.

Unconventional computing currently focuses on synchronization-based computing, neuromorphic and probabilistic computing, and self-organized criticality.

Synchronization-based Computing

HRL is exploring new ways to do energy efficient computing using combinations of novel emerging devices along with computing architectures that allow high accuracy results with low-precision devices. Using arrays of coupled oscillators, it is possible to perform a variety of basic operations commonly found in deep neural networks. These arrays can be constructed from new devices developed from nano structures that can operate with extremely low power. The architectures developed for these novel devices incorporate unique sparsity techniques that can also be applied to conventional CMOS circuits, allowing neural networks for deep learning to be constructed with low-power low-precision elements that rival the performance of their high-precision counterparts. This architecture has led to the development of custom CMOS circuits that are able to achieve over 1.4 Tera-ops per watt. Potential applications for these devices include:

1. Autonomous driving
2. Persistent surveillance
3. Internet of Things
4. Airport security

This material is based upon work supported by the Defense Advanced Research Projects Agency (DARPA) under Contract No. HR0011-13-C-0052. The views expressed are those of the author and do not reflect the official policy or position of the Department of Defense or the U.S. Government.

Neuromorphic and Probabilistic Computing

Neuromorphic computing, with synaptic plasticity, is capable of performing highly scalable Bayesian computation using few resources, accelerating traditional computational problems. Synaptic plasticity can also be used to learn a task through intermittent rewards.

Self-organized Criticality

We have found that computer models of a phenomenon in the brain called self-organized criticality (SOC) can be used to calculate optimal conditions within complex networks. Surprisingly, the search patterns produced by an SOC process have unique properties that lead to more efficient solutions than conventional optimization methods.

Selected Publications:

• H. Hoffmann and D. W. Payton. Optimization by Self-Organized Criticality. Scientific Reports, Feb. 5, 2018