John C. Gallagher

A family of compact genetic algorithms for intrinsic evolvable hardware

For many evolvable hardware applications, small size and power efficiency are critical design considerations. One manner in which significant memory, and thus, power and space savings can be realized in a hardware-based evolutionary algorithm is to represent populations of candidate solutions as probability vectors rather than as sets of bit strings. The compact genetic algorithm (CGA) is a probability vector-based evolutionary algorithm that can be efficiently and elegantly implemented in digital hardware. Unfortunately, the CGA is a very weak, first order, evolutionary algorithm that is unlikely to possess sufficient search power to enable intrinsic evolvable hardware applications. In this paper, we further develop a number of modifications to the basic CGA that significantly improve its search efficacy without substantially increasing the size and complexity of its hardware implementation. The paper provides both benchmark results demonstrating increased efficacy and a conceptual data path/microcontroller design suitable for implementation in digital hardware. Following, it demonstrates efficient implementation by making a head-to-head comparison of field programmable gate array implementations of both the classic CGA and a member of our family of modifications. The paper concludes with a discussion of future research, including several additional extensions that we expect will further increase search efficacy without increasing implementation cost.

The Technology of the Gaps: An Evolvable Hardware Synthesized Oscillator for the Control of a Flapping-Wing Micro Air Vehicle

To date, work in evolvable and adaptive hardware (EAH) has been largely isolated from primary inclusion into larger design processes. Almost without exception, EAH efforts are aimed at creating systems whole cloth, creating drop-in replacements for existing components of a larger design, or creating after-the-fact fixes for designs found to be deficient. This paper will discuss early efforts in integrating EAH methods into the design of a controller for a flapping-wing micro air vehicle (FWMAV). The FWMAV project is extensive, multidisciplinary, and on going. Because EAH methods were in consideration during its earliest design stages, this project provides a rich environment in which to explore means of effectively combining EAH and traditional design methodologies. In addition to providing a concrete EAH design that addresses potential problems with FWMAV flight in a unique way, this paper will also provide a provisional list of EAH design integration principles, drawn from our experiences to date.

The once and future analog alternative: evolvable hardware and analog computation

Once-upon-a-time, analog computers co-existed with their digital counterparts and were considered equally useful. For many applications, specifically equation solving and modeling of physical systems, analog computers were often the better choice. The 1970s, however, saw the beginning of the end of this superiority. Advances in digital circuit fabrication and discrete computer algorithms, not to mention significant advantages of economy, generality, and ease of use, precipitated a mass exodus to general-purpose digital computers so complete that there are now many in the current generation who have neither experience with, nor memory of, the analog alternative. The exodus was certainly made for good reasons. However, it may be beneficial, from time to time, to consider if subsequent developments have rendered those reasons less compelling. This first part of this paper will suggest that the emergence of evolvable hardware (EH) is one such development. It will argue that by applying EH methodologies, one might practically restore the benefits of analog computation as well as achieve benefits not possible in earlier times. The second part of this paper will briefly outline a specific program designed to field practical analog EH control devices.

Evolving neuromorphic flight control for a flapping‐wing mechanical insect

Purpose – The purpose of this paper is to present an approach to employ evolvable hardware concepts, to effectively construct flapping‐wing mechanism controllers for micro robots, with the evolved dynamically complex controllers embedded in a, physically realizable, micro‐scale reconfigurable substrate.Design/methodology/approach – In this paper, a continuous time recurrent neural network (CTRNN)‐evolvable hardware (a neuromorphic variant of evolvable hardware) framework and methodologies are employed in the process of designing the evolution experiments. CTRNN is selected as the neuromorphic reconfigurable substrate with most efficient Minipop Evolutionary Algorithm, configured to drive the evolution process. The uniqueness of the reconfigurable CTRNN substrate preferred for this study is perceived from its universal dynamics approximation capabilities and prospective to realize the same in small area and low power chips, the properties which are very much a basic requirement for flapping‐wing based micr...

An improved evolvable oscillator for all flight mode control of an insect-scale flapping-wing micro air vehicle

In previous work, we presented an adaptive evolvable oscillator that enables online, in-flight, adaptation of a rigorous controller for hovering in an insect-scale flapping-wing micro air vehicle based on the Harvard RoboFly. That particular evolvable hardware oscillator, however, was a proof-of-concept prototype and is incapable of supporting the types of signal adaptation necessary to support on-line correction for other flight modes (E.G. roll, pitch, forward translation, etc.). This paper introduces a new oscillator design capable of supporting signal adaptation for all possible flight modes of the vehicle. It will also present preliminary experimental results demonstrating the adaptive oscillator to be capable of correcting for vehicle faults in a two degree of freedom (2DOF) control task requiring simultaneous regulation of vehicle altitude and roll. The paper will conclude with discussion of application of this adaptive, evolvable oscillator to full vehicle control.

An analysis of the search performance of a mini-population evolutionary algorithm for a robot-locomotion control problem

In this paper, the authors present a performance analysis of a mini-population evolutionary algorithm (EA) on a robot-locomotion control problem. The results indicate that the nature of the search space allows for the design of highly efficient search algorithms that could greatly outperform current hardware-amenable techniques. The authors additionally believe that these search characteristics may be inherent in many practical problems, making the results useful for the larger community.

Evolution and analysis of non-autonomous neural networks for walking: reflexive pattern generators

This paper summarizes the results of the dynamical systems analysis of 422 Continuous-Time Recurrent Neural Network (CTRNN) single-leg locomotion controllers evolved under conditions where reliable sensory information was always available. The general principles underlying the operation of all 422 resulting Reflexive Pattern Generators (RPGs) are discussed. Several RPG operation features are explained and verified. Finally, discussion is made of future extensions of this research.

An Improved Evolvable Oscillator and Basis Function Set for Control of an Insect-Scale Flapping-Wing Micro Air Vehicle

This paper introduces an improved evolvable and adaptive hardware oscillator design capable of supporting adaptation intended to restore control precision in damaged or imperfectly manufactured insect-scale flapping-wing micro air vehicles. It will also present preliminary experimental results demonstrating that previously used basis function sets may have been too large and that significantly improved learning times may be achieved by judiciously culling the oscillator search space. The paper will conclude with a discussion of the application of this adaptive, evolvable oscillator to full vehicle control as well as the consideration of longer term goals and requirements.

Improvements to the *CGA enabling online intrinsic evolution in compact EH devices

Recently, we proposed a neuromorphic intrinsic online evolvable hardware (EH) system designed to learn control laws of physical devices. Since we intend to eventually build this device using mixed signal VLSI techniques, and because we intend to address control applications in which small size and low power consumption are critical, we are extremely concerned with the design of physically compact devices. This paper focuses on the evolutionary algorithm (EA) portion of our proposed system. We discuss modifications to our previously reported *CGA that significantly increases its performance against dynamic optimization problems without significantly increasing the amount of hardware required for implementation. We demonstrate the efficacy of our improvement by testing against two series of moving peak benchmarks. We conclude with discussions of both the implications of our findings and our plans for future work.

An Islands-of-Fitness Compact Genetic Algorithm Approach to Improving Learning Time in Swarms of Flapping-Wing Micro Air Vehicles

Insect-Scale Flapping-Wing Micro-Air Vehicles (FW-MAVs) may be particularly sensitive to degradation of pose and position control caused by ongoing or pre-existing damage to the airframes. Previous work demonstrated that in-flight recovery of sufficient pose and position control precision via use of an adaptive oscillator component inside traditional SISO controllers. This work will replace previously used oscillator learning algorithms with a hyperplane sampling Evolutionary Algorithm (EA) that employs cross-vehicle islands-of-fitness. It will be demonstrated that this strategy allows swarms of vehicles to cooperatively, and more quickly, find and correct for simulate manufacturing errors that appear in all vehicles – even in the presence of randomized vehicle specific errors that are not common to all vehicles in the swarm. The paper will present specific simulation results demonstrating efficacy of this scheme and discussion of future applications of islands-of-fitness methods in this problem domain.