Encoder Front Page
SRS Home | Front Page | Monthly Issue | Index
Google
Search WWW Search seattlerobotics.org

ANCOB
Artificial Nervous COntrolled Beetle

Design & Development

George Vastianos
Electronics Engineer BSc.
Dipl. from Electronics Department,
Faculty of Technological Applications,
Technological Educational Institute of Piraeus, Greece

www.vastianos.com

Table of contents

0. Abstract
1. Chaos theory
1.1 Applications of chaos theory
1.2 Chaotic oscillators
2. Nervous networks
2.1 Central Pattern Generators
3. Biomorphic control systems
3.1 Quasi-periodic nervous neuron
4. BEAM robotics philosophy
5. Beetle anatomy
6. Implementation of ANCOB robot
6.1 Hardware implementation of brain's Nv network
6.2 Robot's photo gallery
7. References

Keywords: Artificial intelligence & life, neuroscience, nervous networks, biomorphs, mobile robotic systems, bio-robotics, micro-robotics, BEAM-robotics.

0. Abstract

The idea of building machines that reflect biological structures is certainly not new. From Leonardo Da Vinci who dreamt of winged flying machines to the clockwork automata, humans have been trying to build machines that can perform tasks as successfully as the myriad creatures in our natural world. ANCOB is a micro-robot that uses a tiny nervous network (Nv) to mimic, in a general way, the living of a real-biological beetle. ANCOB’s brain kernel consisted of only two Nv neurons that have been constructed using four transistors and some other simple analog electronic components. ANCOB’s brain is not more than a double quasi-chaotic oscillator used for the development of a robotic creature that is minimalistic in design but complex in behaviors…

1. Chaos theory

Chaos theory is among the youngest of the sciences, and has rocketed from its obscure roots in the seventies to become one of the most fascinating fields in existence. At the forefront of much research on physical systems, and already being implemented in fields covering as diverse matter as arrhythmic pacemakers, image compression, and fluid dynamics, chaos science promises to continue to yield absorbing scientific information which may shape the face of science in the future.

Formally, chaos theory is defined as the study of complex nonlinear dynamic systems. "Complex" implies just that, "nonlinear" implies recursion and higher mathematical algorithms, and "dynamic" implies non-constant and non-periodic. Thus chaos theory is, very generally, the study of forever changing complex systems based on mathematical concepts of recursion, whether in the form of a recursive process or a set of differential equations modeling a physical system.

Chaos theory has received some attention, beginning with its popularity in movies such as "Jurassic Park"; public awareness of a science of chaos has been steadily increasing. However, as with any media covered item, many misconceptions have arisen concerning chaos theory. The most commonly held misconception about chaos theory is that chaos theory is about disorder. Nothing could be further from the truth! Chaos theory is not about disorder! It does not disprove determinism or dictate that ordered systems are impossible; it does not invalidate experimental evidence or claim that modeling complex systems is useless. The "chaos" in chaos theory is order - not simply order, but the very ESSENCE of order.

It is true that chaos theory dictates that minor changes can cause huge fluctuations. But one of the central concepts of chaos theory is that while it is impossible to exactly predict the state of a system, it is generally quite possible, even easy, to model the overall behavior of a system. Thus, chaos theory lays emphasis not on the disorder of the system - the inherent unpredictability of a system - but on the order inherent in the system - the universal behavior of similar systems. Thus, it is incorrect to say that chaos theory is about disorder. To take an example, consider Lorenz's Attractor. The Lorenz Attractor is based on three differential equations, three constants, and three initial conditions.


Lorenz Attractor - XZ Plane Projection

The attractor represents the behavior of gas at any given time, and its condition at any given time depends upon its condition at a previous time. If the initial conditions are changed by even a tiny amount, say as tiny as the inverse of Avogadro's number (a heinously small number with an order of 1E-24), checking the attractor at a later time will yield numbers totally different. This is because small differences will propagate themselves recursively until numbers are entirely dissimilar to the original system with the original initial conditions.

However, the plot of the attractor will look very much the same. Both systems will have totally different values at any given time, and yet the plot of the attractor - the overall behavior of the system - will be the same. Chaos theory predicts that complex nonlinear systems are inherently unpredictable - but, at the same time, chaos theory also insures that often, the way to express such an unpredictable system lies not in exact equations, but in representations of the behavior of a system - in plots of strange attractors or in fractals. Thus, chaos theory, which many think is about unpredictability, is at the same time about predictability in even the most unstable systems.

1.1 Applications of chaos theory

Everyone always wants to know one thing about new discoveries - what good are they? So what good is chaos theory? First and foremost, chaos theory is a "theory". As such, much of it is of use more as scientific background than as direct applicable knowledge. Chaos theory is great as a way of looking at events which happen in the world differently from the more traditional strictly deterministic view which has dominated science from Newtonian times. Moviegoers who watched "Jurassic Park" are surely aware that chaos theory can profoundly affect the way someone thinks about the world; and indeed, chaos theory is useful as a tool with which to interpret scientific data in new ways. Instead of a traditional X-Y plot, scientists can now interpret phase-space diagrams which - rather than describing the exact position of some variable with respect to time - represents the overall behavior of a system. Instead of looking for strict equations conforming to statistical data, we can now look for dynamic systems with behavior similar in nature to the statistical data - systems, that is, with similar attractors. Chaos theory provides a sound framework with which to develop scientific knowledge.

However, this is not to say that chaos theory has no applications in real life. Chaos theory techniques have been used to model biological systems, which are of course some of the most chaotic systems imaginable. Systems of dynamic equations have been used to model everything from population growth to epidemics to arrhythmic heart palpitations. In fact, almost any chaotic system can be readily modeled - the stock market provides trends which can be analyzed with strange attractors more readily than with conventional explicit equations; a dripping faucet seems random to the untrained ear, but when plotted as a strange attractor, reveals an eerie order unexpected by conventional means.

Fractals have cropped up everywhere, most notably in graphic applications like the highly successful Fractal Design Painter series of products. Fractal image compression techniques are still under research, but promise such amazing results as 600:1 graphic compression ratios. The movie special effects industry would have much less realistic clouds, rocks, and shadows without fractal graphic technology. And of course, chaos theory gives people a wonderfully interesting way to become more interested in mathematics, one of the more unpopular pursuits of the day.

1.2 Chaotic oscillators

At present, active search for possible chaos applications goes. However to apply dynamical chaos, it is necessary to have signal sources generating chaotic signals. As is known, chaotic oscillators play the role of the sources.

Now, there is a large variety of chaotic oscillators which differ from each other by both the structure and elements, and which have various characteristics. On the other hand, the use of chaotic signals in some applications is possible when oscillators have specific characteristics. A concrete shape of power spectrum is one of them.

As is known, there is an approach to design of chaotic oscillators with preassigned band and shape of spectral characteristics of the output signal based on ring structure oscillating systems. The basic concept of the approach is to introduce elements into the oscillating system feedback loop which, on the one hand, provide conditions for chaotic generation and, on the other hand, form required spectral characteristics.

2. Nervous networks

Much of control systems and systems theory is derived from mathematics such as Lyapunov's Stability Theory. The approach of Mark W. Tilden (inventor of the BEAM robotics philosophy) differs in that it is motivated by biological neurons/nervous systems. Many neural systems in nature are surprisingly adaptive and efficient. Taking advantage of the solutions nature provides, researchers such as Hopfield are building analog models of specific nervous system functions.

An explanation of the elementary function of artificial neurons is useful for the development of the Nv neuron model. The human nervous system is an analog system which consists of approximately ten trillion neurons, each having the ability to receive, process, and transmit electrochemical signals over complex pathways. Neurons are connected by dendrites, which extend from cell body to cell body. Signals are received at a connection point called synapse and sent to the cell body. The inputs are summed. The input either excites or inhibits the cell. If the cumulative excitation in the cell body exceeds a threshold, the cell fires, sending the signal to the other neurons. The Nv neuron and most artificial neural networks model these simple neuron characteristics.

2.1 Central Pattern Generators

Oscillatory neural networks are groups of neurons responsible for a wide variety of periodic behavior patterns such as locomotion, breathing, and chewing. The study of oscillatory neural networks in invertebrates accounts for much of what is known about this rhythmic behavior. Documented examples include the digestive rhythms of the lobster, stepping movements of the cockroach, and the rapid wing motion of the locust during flight. In general, these types of networks are capable of generating oscillatory activity without requiring some sensory input, although some do require a form of initial excitation. Such networks are referred to as central pattern generators.

Neuronal oscillators could work in two ways: (1) one or more cells embedded in a network have the property of bursting (cell-driven oscillators) or (2) the network itself produces bursts as a result of synaptic interactions. Several combinations of such networks exist. A hybrid form in which the oscillatory pattern is generated by both bursty cells and network interactions (mixed oscillators) is the predominant form. These networks consist mainly of inhibitory synapses.

Some CPGs are capable of producing multiple patterns of activity. The small, localized CPGs, which occur in invertebrates, make it possible to study the relationship between the emergent collective behavior of the biological network and the network's underlying circuitry. Much has been done to analyze the dynamical properties of invertebrate CPGs. Cellular, synaptic, and connectivity properties have been experimentally manipulated. Cell interactions within the network have been simulated and analytical equations describing the network dynamics have been developed. There is no precise explanation for the mechanisms that control frequency, duration, and phase relations of the motor pattern. Studies of invertebrate CPGs such Getting's network simulation of the escape swimming rhythm of the mollusc show that the generation of these rhythms is a complicated process involving the influence of multiple modulators which modify the output of the circuit.

The architectures (or topologies) of connectionists models for artificial neural networks are not biologically realistic. Neurons have complex physiological properties that are important to their computation and time dependent properties that are ignored in most models. An exception is a computational model in which the sizes of variables are represented by the explicit times at which action potential (neuronal firing) occurs. Analog information is represented by using the timing of action potentials with respect to a continuous collective oscillatory pattern. Experiments on rat hippocampal "place cells" and electric fish show this action-potential phase-time coding.

3. Biomorphic control systems

Biomorphic control (from the Greek word biology "to live" and morphology "to take form") systems place an equal importance on mechanical and electrical structure. The structures are an integral part of one another and a deficiency in either hierarchy yields a less successful, less capable device. A biomorphic, autonomous system attempts to (1) achieve directed motion in the environment, (2) procure abundant power, and (3) robustly (even aggressively) negotiate obstacles.

The autonomy of these robots is inherent to their structure. The robot has no knowledge of the surrounding environment and its motion is not controlled by a microprocessor. Rather, an analog computation of the sensory input and the internal process state produces a directed response. The tenuous relationship between weight to power, mass imbalances, coupling and biasing the neuron for dynamic motion are the challenges of designing a biomorhpic system.

3.1 Quasi-periodic nervous neuron

The basic circuit used in minimalist "Biomechs" is the quasi-periodic nervous neuron (Nv) shown below:

The input of the neuron is pin #1 and pin #2 (where #1 = +V, #2 = 0V) and the output is pin #3 and pin #4 (where #3 = +Vload, #4 = -Vload). The input of the neuron can be connected with a low current - constant voltage power supply source or a low current - dynamically variable voltage power supply source (like a solar cell). The output of the neuron can be connected with any low resistance load (like an inductive load). Most of the times the load is a small motor.

Also instead of D1, any electronic part can be used if its resistance according voltage is non-linear in such a way that there is a voltage level where over this level the resistance decreases dramatically (Vtriggering). This triggering voltage level can be constant or can be changed dynamically.

The quasi-periodic nervous neuron can be considered an effective quasi-chaotic oscillator, more so when considering variables in motor load, inertia, variability in triggering voltage level or variability in power supply voltage.The advantages of this design are small component count and adjustability, but mostly its very low current drain until tripped. This means that Biomorphic designs can be very small, robust, and self-contained. Coupled clusters of these oscillators provide the dynamical richness of Biomech systems.

The first quasi-periodic nervous neuron was built by Mark W. Tilden in late 1989.

Theory of operation

Essentially the quasi-periodic nervous neuron is a modified SCR (Silicon Controlled Rectifier) with supercritical feedback. I would like to point out that we are interested in the electron current, as opposed to conventional current (that is, the movement of positive charges, opposite to electron current), unless otherwise stated as so.

The following timeline are the events and circuit characteristics while charging and discharging of C1:

4. BEAM robotics philosophy

BEAM is the brainchild of Mark W. Tilden who is currently working as a researcher at Los Alamos National Laboratory, USA. BEAM is an acronym standing for Biology, Electronics, Aesthetics, Mechanics:

Biology: It's tough to beat 4 billion years of evolution; the world around us is a wonderful source of inspiration and education. Bear in mind, of course, that unlike Mother Nature, you also have the advantage of gears, motors, bearings, and good glues!
Electronics: It kind of goes without saying, but this is what you'll use to drive your creations. BEAM robotics, though, strives for rich behaviors from simple circuits. Here's the key: simple and understandable circuits, surprisingly complex in behavior.
Aesthetics: This just means your creations should look good. I'm an engineer, but even I appreciate a good-looking design. Besides, if a design looks "clean," it's more likely to work (and easier to test / debug) than a design that's tangled and unruly.
Mechanics: This is the less-than-obvious secret of many successful BEAMbots -- with a clever mechanical design, you can reduce the complexity of the rest of your robot (reducing the number of motors and sensors, for example).

BEAM robotics basically starts from 3 philosophical tenets:

Use minimalist electronics: This keeps complexity from "snowballing", and keeps costs down.
Recycle & reuse components out of technoscrap: This keeps things cheap, and avoids a lot of trips to parts stores; virtually all the parts required to make a BEAM robot can be found in broken electronics (ovens, walkman's, CD players, VCRs, pagers...).
Solar power your critter if possible: While less powerful than even a small battery (and, up-front, more expensive), solar cells last for years; solar-powered BEAMbots don't require constant battery replacements or down-time for battery recharging.

5. Beetle anatomy

Beetle is a general term for insects of the order Coleoptera. There are known to be over 370,000 species of beetle, outnumbering all the known species of vascular plants, and six species of beetle for every one vertebrate, with an estimated five million more species yet to be discovered.

Beetles undergo a complete metamorphosis, with a distinct pupal stage intervening between life as a larva and a sexually mature adult. As with other insects of this type, the larvae stage represents the principal feeding stage. Most adult beetles have a robust, hard external skeleton (carapace) which acts like body-armour and a pair of horny wing-cases (elytra), which usually completely cover the hind part of the body including the abdomen.

Even if there are so many different species of beetle the main anatomy structure remains the same. So, abdomen is the segmented tail area of a beetle that contains the heart, reproductive organs, and most of the digestive system. The maxillary palps are long, segmented mouth parts that grasp the food. The thorax is the middle area of the beetle's body - where the 6 jointed legs and wings are attached. Elytra are the hardened fore wings that protect the longer hind wings. Also beetles have two hind wings, used for flying (or swimming). These long wings can be folded under the elytra when not in use. The head is at the front end of the beetle's body and is the location of the brain, the two compound eyes, the mouth parts, the pharynx (the start of the digestive system), and the points of attachment of its two antennae.

Beetles display a wide variety of life-styles and behaviour according their type and the environmental conditions but the most common characteristic in their behavior is the preference about light conditions. Beetles, like cockroaches, characterized from their photophobic behavior, something that makes them to hate light and run away from it.

6. Implementation of ANCOB robot

The goal of this project was the creation of a biomorphic robot that will work in a general way as an artificial beetle. Because of this, the design of this robot focused mainly to the definition of its behaviors and the external shape (even if it is very difficult to construct an electronic creature that will look like a real beetle).

The following list presents the three behaviors of the presented robotic beetle according to the priority:

All the above-described behaviors finally implemented very well by using the non-linear characteristics of the transistors of a quasi-periodic nervous neuron (Nv) pair. You can understand how this happens (in practice) if you study carefully the circuit of the next section.

6.1 Hardware implementation of brain's Nv network

Finally the implementation of the robot’s brain based on an Nv network that consisted of two quasi-periodic nervous neurons. Phototransistors have been used as eye sensors and small dc motors as leg actuators. The electronic circuit that implements the Nv network is the following:

PARTDESCRIPTION
PTR1,2VPT50A3
TR1,32N3906
TR2,42N3904
R1,222R/0,5W
R3,42K2/0,25W
C1,24700uF/25V
M1,2DC MOTORS (MABUCHI RF-300C-14270)
FLED1,2RED FLASHING LEDS
BT1,21,5V AA BATTERIES

As you see above, the two quasi-periodic nervous neurons are connected in such a way that they have a common input but two independent outputs. The input of each neuron is connected with the same low current - constant voltage power supply source (BT1,2 and R1,2) and the output of each neuron is connected with an inductive load -motor (M1,2). The triggering voltage of each neuron changes dynamically according to the luminance that has been enforced to its phototransistor (PTR1,2). As result we have a double quasi-chaotic oscillator that the frequency of each output is completely unpredictable and depends from variables in motor load, inertia, variability in triggering voltage level or variability in power supply voltage.

When the circuit starts to work then a small current (that caused because of R1,2) starts to charge the capacitors (C1,2). The triggering voltage of each oscillator is defined by its phototransistor and its flashing LED. The flashing LEDs are used to provide a constant minimum triggering voltage to the oscillator and the phototransistors are used to increase dynamically this triggering voltage according to the enforced light. So when the enforced light is enough to cause a triggering voltage level lower than the power supply voltage level then the specific neuron will trigger at this voltage level otherwise if this level is higher than the power supply voltage then the neuron will never trigger (REST behavior). If the power supply voltage becomes lower than the minimum triggering voltage then the neuron stops to trigger at all (LIVE behavior). If the light conditions and the power supply voltage are able to make the neurons to trigger then the neuron that will trigger first is the one with the lower triggering voltage level (higher luminance). When a neuron triggers the result is the discharging of the capacitors through its load. This means that we can have only one triggering at a time and by using the motors (loads) for locomotion in a correct way we can make this system to change its position to a position with lower luminance (SEEK behavior).

Finally the use of this double oscillator with the unpredictable output frequencies proved useful for the development of a robotic creature that is minimalistic in design but complex in behaviors…

6.2 Robot's photo gallery

Front view Side view Top view
Bottom view Brain view

7. References

  1. Vastianos G., "ANCOB - Artificial Nervous COntrolled Beetle", (Draft Notes), (12/2001).
  2. Ho A., "Chaos Introduction", (27/10/1995).
  3. Maximov N., Panas A. and Starkov S., "Chaotic oscillators design with preassigned spectral characteristics", ECCTD ’01 - European Conference on Circuit Theory and Design, August 28-31, 2001, Espoo, Finland.
  4. Dmitriev A., Panas A. and Starkov S., "Ring oscillating systems and their application to the synthesis of chaos generators", Int. J. of Bifurcation and Chaos, Vol. 6, No. 5, 1996, pp. 851-865.
  5. Frigo J. and Tilden Mark W., "SATBOT I: prototype of a biomorphic autonomous spacecraft", Los Alamos National Laboratory, Los Alamos, New Mexico, USA, (27/10/1995).
  6. Hasslacher B. and Tilden Mark W., "Living Machines", Robotics and Autonomous Systems: The Biology and Technology of Intelligent Autonomous Agents, Elsivier Publishers, Spring 1995.
  7. Hopfield J.J., "Pattern recognition computing using action potential timing for stimulus representation", Vol. 376, pp. 33-36, 1995.
  8. Wasserman Philip D., "Neural Computing Theory and Practice", Chp. 1-2, Van Hostrand Reinhold, New York., 1989.
  9. Bower James M. and Beeman D., "The Book of GENESIS", Chp. 8, TELOS, Santa Clara, 1995.
  10. Selverston Alien L. and Moulins M., "Oscillatory Nerual Networks", Vol. 47, pp. 29-4S, Annual Reviews Inc., 1985.
  11. Tilden Mark W., "The Design of Living Biomech Machines: How low can one go?", Los Alamos National Laboratory, Los Alamos, New Mexico, USA, (07/1997).
  12. Tilden Mark W. and Hasslacher B., "Theoretical Foundations for Nervous Nets and the Design of Living Machines", Los Alamos National Laboratory, Los Alamos, New Mexico, USA, (10/1995).
  13. Bush O.B., "BEAM - Robotics - Tek: The Solar Engine", (02/12/1998).
  14. Seale E., "A bit of background on BEAM: BEAM - what it is, what it's used for.", (02/03/2002).
  15. Enchanted Learning, "The Beetle Anatomy Diagram", (2001).