Orbital Research Neural Networks For Control and Signal Processing User Manual

nimals must integrate large amounts of sensor information to orient movement

towards goals and away from threats, often in milliseconds, to survive. Nature,

through evolution and natural selection, has optimized this behavior. In particular,

insects initiate actions reflexively in situations where no time for detailed information

processing or planned decision making is possible. Orbital Research, Inc. has

developed a neural circuit based upon over twenty years of research by biologists

Biologically inspired neural circuits for

autonomy and sensor fusion

on the escape response of the American cockroach that provides a key tool for

achieving this capability.

Biological Inspiration

Alt houg h “Biologically Inspired” approaches are becoming increa singly popular in engin eering, few research groups have succeeded in tightly integrating biological principles into co ntroller design. Tight int egration of biological principles requires an appro ach invol ving close feedback between research in biology and engineering. Key steps for ma king this tightly integrat ed approach are:

Identific ation of behavioral cap abilities of biological organi sms directly relevant to the task of interest

Identification of biological species that are suited for studying these capabiliti es

Drawing upon biological studies and experimental evidence to understand how natural systems accomplish desired tasks

Abstract and implement those capabilities that are needed for a particular engin eering task.

The neural circ uit is based upon the work of

biologists studying the American Cockroach.

Multi Constraint Incorporation:

higher-level path planning behavior where the vehicular dynamics and constraints (such as varying flight envelopes) are not considered. Autonomous vehicles with this biological solution are capable of mission optimization within a host of varying conditions.

Instantaneous Path Generation and Tracking Reactions:

control methodologies are capable of extremely rapid reactions. Nat ure ha s addressed thi s need through the develo pment of locally controlled reflexes as in the American Cockroach whose neural organization is mimicked for autonomous vehicle control. Thus, auto nomous vehicles with this biologically inspired algorithm are capable of processing, planning around, and reacting to changes or threats in real-time .

Evo lved Pa tt ern Resp on se :

is the result of the inco rpora tion of millions of generations of natural selection. T his permits robust instantaneous reactions to every situation by combining and tuning pre-developed reflexes. Thus, autonomous vehicles can rapidly reconfigure operation al scenarios to accomplish the overall mission goal.

Many autorouters only address

Few

The response of bi ological organis ms

By working closely with biologists under this paradigm, ORI has developed several reflex control algorithms that incorporate many of the advantages seen in biological organisms, including:

Sensor Integration :

large amount s of sensory informat ion from multiple sensors and sensor types so action decisions may be made rapidly. Aut ono mous vehic les with this capabi lity reac t to current f ligh t conditions and mission scenarios in real-time.

Context Depend ent Be havior:

continuously updated based upon physiological state and environment. Enabling this ability in autonomous vehicles increases their efficiency and adaptability.

Orbital Research, Inc.

4415 Euclid Ave., Suite 500

leveland, OH 44103-3733C

In na ture, animals are capable of in tegrating

Animal's react ions are

Implementation of cockroach escape reflex4es as an obstacle avoidance system for autonomous vehicles, the cockroach’ s neural architecture is labeled with italics and the obstacle avoidance system’s analogs are shown in the blocks

Contact: Frederick J. Lisy, Ph.D.

Te lephone (216) 649-0399

E-mail lis y@orbitalresearch.com

www.orbitalresearch.com

Rev: RMK 2-2-2004

Neural Net reflexes enable robust threat response as well as targeting for Unmanned Air Vehicles

Autonomous Threat Avoidance and Targeting

Orbital Research Inc. (ORI) has developed a revolutionary set of algorithms that produce rapid threat avoidance and target seeking reflexes for autonomous vehicles su ch as U nm anned Air Vehicles (UAV) or Unmanned Ground Vehicles(UGV). The reflex provides near instan taneo us, context dependent integrat ion of sensor data and dynamic path rep lanning for autonomous vehicles by mimicking the processing capabilities of biological organisms. This work was originally generated at Case Western Reserve University (CWRU), which invested 20 years into studying the biological and evo luti onary d evelo pment of biological org anisms. Orbital Research has transitioned the first implementation of this technology to autonomous vehicles for target seeking (BioSeek) and threat/obstacle avoid ance (Bio AVERT). In both simulation and demonstration flights using an unmanned air vehicle, ORI has proven t he efficacy of these artificial reflexes:

Target seeking reflex sh own reacting to a detected virtual target dur ing flight testing.

Sensor Fusion for Next Generation Navigation

Researchers at ORI are currently developing integrative architectures for sensor fusion that take advantage of the sensor fusive capabilities and context dependent behavior of the cockroach escape response. The unique ability to combine contextual awareness with sensor fusion offers tremendous potential for the development of s ystems that in corporate informat ion from di vers e sources to produce coherent and accurate data. One implementation currently under deve lopment is an Ultra Tightly Cou pled GPS/INS system that integrates GPS signal information with the raw data from Inertial Measurement Units (IMU) to simultaneously provide robust and accurate position data and track the GPS signal. It is believed that this system will provide increased precision and accuracy while proving more robust to sensor misalignment, GPS si gnal drop out and GPS signal jamming or spoofing. In addition, this technology has application to many in the other sensor fusion problems such as those occuring in the use of radar, sonar, and phase and focal plane arrays of sensors.