Integration of Cellular Biological StructuresInto Robotic Systems

J. Gough, G. Jones, C. Lovell, P. Macey, H. Morgan,F. D. Revilla, R. Spanton, S. Tsuda, and K.-P. Zauner∗

1 Introduction

Ideas from biological information process-ing have influenced numerous robot controlarchitectures. Most of them draw their in-spiration from neural networks. Many cog-nitive capabilities can only be understoodfrom a system level perspective that in-tegrates body, control, and envrionment.This insight led to robot designs that ex-ploit the physical characteristics of a robots’body [1]. If one takes the direction of theseinvestigations further, one arrives at thepoint where control is not only exploitingthe physics of the body, but is in itself di-rectly driven by physics.

Arguably this step is necessary to narrowthe performance gap between man-madedevices and biological systems. All organ-isms require information processing to de-fend their intricate organisation against theonslaught physical entropy. As a conse-quence organisms exhibit an intriguing so-phistication in overcoming computationallydifficult challenges. This is the case evenfor simple life forms and individual cells.

∗School of Electronics and Computer Sci-ence, University of Southampton, Southamp-ton SO17 1BJ, United Kingdom; contact:kpz@ecs.soton.ac.uk

Robots and organisms both need to processsensory information from a range of modal-ities and make decisions based on noisyambiguous data. Not only do they needto process the sensory inputs and respondin real-time with appropriate actions, butthey need to do this under the resourcerestrictions imposed by their size. Organ-isms approach this challenge by directly ex-ploiting the physico-chemico properties oftheir molecular materials—in stark contrastto present computing technology the for-malisms of which are defined in disregardof the physical substrate used to imple-ment. Brooks observed that ‘the matterthat makes up living systems obeys the lawsof physics in ways that are expensive to sim-ulate computationally” [2] and Conrad ar-gued that programmable computing is nec-essarily inefficient compared to nature’s in-formation processing [3, 4]. From what hasbeen stated above, it appears worthwhileto give more attention to the role computa-tional substrates may play in cognitive sys-tems [5]. Taking heed of the clues from biol-ogy, macromolecular materials appear par-ticularly attractive [6].

1

Figure 1: A tethered hexapod robot op-tically interfaced with a Physarum poly-cephalum cell [7].

2 On-board Cellular

Robot Control

We endeavour to sketch a potential routeto recruiting some of nature’s informa-tion processing efficiency for technical ap-plications: the integration of living cellsinto bio-electronic hybrid robot control de-vices. In this approach, selected featuresof the robot’s environment are mappedinto features of the environment of a liv-ing cell. Conversely, the response of thecell is mapped back onto the actuators ofthe robot. Fig. 1 shows an early implemen-tation of this concept. The cell employedin the experiments is the plasmodial stage

of the slime mould Physarum polycephalum.This large multinucleated cell can be foundon decaying wood and feeds on bacteria andorganic matter. Despite its macroscopicsize (it typically grows to >100 cm2) the cellacts as a single integrated organism. Toolarge to rely on diffusion, it uses rythmiccontractions to distribute materials withinits cell body [8, 9].

Central to our approach is the cell-robotinterface. Both, optical and electronic in-terfaces for plasmodia of the slime moldhave been implemented. Our first approachto cellular robot control had purposely setaside the question of whether it is practica-ble to integrate a cell-based controller intoa small robot [10]. In recent work we haveaddressed this issue. The current architec-ture uses the oscillator design suggested byTakamatsu and Fujii [11] but with a maskon which electrodes have been patterned.The mechanical oscillations of the plasmod-ium that were previously detected as lo-cal changes in light transmission of the cellbody, can also be detected by the change inimpedance among the electrodes integratedin the mask [12, 13].

Fig. 2 shows a mask with two indepen-dent plasmodia (labeled Physarum in thefigure). Each is grown into a cutout in thecircuit board that acts as mask to confinethe plasmodium into a dumbbell shape witha left and a right well connected by a chan-nel. Two pairs of electrodes are availablefor each well [11]. The image shows the un-derside of the chip through a cutout in thecircuit board with the circuit for measuringthe impedance. During operation, this faceof the mask would be covered with a layerof agar. To monitor the oscillations of theslime mould cell, we employ an integrated

2

Physarum

5 mm

Figure 2: Electrical interfacing of thePhysarum plasmodium.

circuit capable of measuring impedance atup to 100 kHz. For further details refer tothe Methods section.

The impedance across each of the eightpairs of electrodes is measured once persecond, the maximal rate our multiplexedsetup permits. Fig. 3 shows the signals orig-inating from two wells, i.e., two ends of asingle cell, after a moving average filter hasbeen applied to reduce noise.

The two curves show the moving averagesover 15 samples, recorded from the rightand left well of the plasmodium. The phase-relationship between the two wells is shownin the bar across the bottom of the figurewhere white and black indicate in- and out-of-phase, respectively. In the experimentdescribed here, this binary phase-relation isused to steer a robot. The minimalist robot,pictured in Fig. 4, was inspired by Braiten-berg’s vehicles [14] and is designed as a testplatform for unconventional controllers.

Figure 3: Change in impedance at 100 kHzconcomitant with the volume oscillations oftwo parts of a single slime mould cell.

Impedance Circuitry Physarum

Constraint sensor Data processing

Figure 4: Robotic platform with thePhysarum chip installed on the impedancecircuit board.

3

In the present experiments a simple one bitactuator control patterned on the control ofactuation in bacterial chemotaxis [15, 16]has been employed. The control bit tog-gles between the state of turning in a ran-dom direction or moving forward. The stateis updated at a rate of 1 Hz, whereby in-phase oscillation of the two wells mappedto random turning and out-of phase oscil-lation is mapped to straight moves. A typ-ical segment of a trajectory without stim-ulation is shown in Fig. 5. An illuminatedtarget on the robot is tracked for positionand orientation with an overhead camera.If the robot reaches the boundary of thearena, forward commands are ignored andonly turns are executed.

So far this prototype only implements theactuator side; the facility to transmit thesensory stimulation to the slime mould isnot yet completed. Nevertheless, prelimi-nary tests indicate that in principle the in-tegration of living cells into a robot is fea-sible.

3 Vision: Living Devices

Microbial cells are routinely employed inmany engineered processes. Their ability toself-reproduce offers a cheap and fast routeto deploy complex nano-scale systems. Par-allel to the progress in molecular biology,the traditional breeding (e.g., to optimisebaker’s yeast) has been augmented with di-rect modification of cell lines to introducenew capabilities (e.g., production of humaninsulin). What started as metabolic engi-neering with static changes to the genome of

A

B

Figure 5: The arena to which the robot isconfined in the view of the tracking camera(A), and the trajectory of the robot corre-sponding to the signals shown in Fig. 3; ◦marks the start and • the end of the seg-ment (B).

4

the cells, is now, as “synthetic biology,” onthe way to engineer dynamic control struc-tures and may eventually allow the design ofapplication specific cells [17]. In addition toself-reproduction, cells provide quality con-trol and testing at the point of assembly formany of their macromolecular products andcan replenish their components and thusoperate under adverse conditions.

Another important development is the in-creasing sophistication of lab-on-chip andmicrofluidic technologies that make it pos-sible to maintain a controlled microenviron-ment for a cell. The confluence of bothof these lines of research opens up a newengineering direction where living cells aretightly integrated with conventional tech-nology. In the resulting devices, living cellsare an essential component. Examples are“living sensors” [18, 19] and “living pumps”[20, 21]. In the coming decades such an ap-proach will almost certainly be explored forelementary cognitive systems and we see theongoing work described in the previous sec-tion as a small step in this direction.

4 Potential for Space

Applications

The key features of cells that may be of po-tential interest in the space field are:

1. Self-repair capability

2. Self-reconfiguration capability

3. Efficient in material requirement

4. Efficient in energy usage

Cells, in general, will recognise a wide va-riety of damages and take steps to re-pair themselves. Some extremophile mi-crobes, in particular, posses elaborate in-nate mechanisms to actively maintain theirmolecular organisation. The material ef-ficiency of cells usually permits redun-dancy of functional components (especiallyproteins) and information carriers (doublestranded DNA) and thus enables them toautonomously recover from many injuries.A slime mould plasmodium can containmany millions of redundant copies of its ge-netic information within a single cell. Po-tentially cells could offer a unique solutionto the problem of transporting a complexfunctional system over a long period thougha harsh environment when shielding is im-practical due to weight constraints. It isalso possible to imagine applications of cellsto grow a functionalised surface such as asensory-skin for a robot. Only a small vol-ume of seed cells (or spores) would need tobe protected from radiation damage duringtransit and after formation such a surfacemay be able to self-heal when damaged.

Of course there are also problems. De-vices containing living cells pose the dan-ger of spreading organisms from Earth ormay interfere with the search for indicatorsof non-terrestrial life. Cells do have differ-ent requirements from inanimate materials,have a quite restricted operating tempera-ture range, need to be protected against lossof moisture, and in general have idiosyn-cratic supply needs.

5

5 Conclusion

Robots and organisms face the same prob-lem: they need to act in real-time in acomplex environment and are severely re-stricted in material and energy that canbe allocated to their information process-ing needs. Organisms, however, spectacu-larly outperform current robots. It appearsincreasingly likely that the choice of compu-tational substrate is critical to their success.

The integration of living cells into roboticsystems is feasible, albeit at its infancy.Cells offer many properties that are un-likely to come within reach of conven-tional technologies within the foreseeablefuture, in particular self-reproduction andself-repair of complex heterogeneous struc-tures. Devising bio-hybrid architectures inwhich cells confer their unique characteris-tics in a beneficial manner to the system asa whole, will be among the key challengesin this field.

Conceivable early application domains inspace research may lie in non-critical sub-systems where the integration of living de-vices potentially offers the opportunity todeploy systems of high complexity with areasonable chance to recover, or grow func-tionality after extended periods of exposureto high electromagnetic radiation flux. Itgoes without saying that any applicationscenario needs to exclude the possibility ofcontaminating celestial bodies.

Methods

Electrodes were patterned on a small circuit board(PCB); the wells drilled (cf., Fig. 2) and a channelcut between each pair of wells. The design of theelectrode pattern allows for two independent plas-modia. Each plasmodium is in a dumbbell configu-

ration with two 1.6 mm diameter wells at a centre-distance of 2.5 mm and connected by a 0.4 mmwide channel. Two pairs of electrodes are avail-able on each well for a total of 16 electrodes onthe board. The copper side of the circuit board islaminated except at the areas where it connects tothe impedance circuit. To provide the plasmodiumwith a moist surface the board is placed with thelaminated side on layer on agar. Physarum poly-cephalum was cultured on 1.5% agar gel and fedwith oat flakes. It is possible to let the slime mouldgrow onto the agar surface masked with the cir-cuit board. However, to accelerate the preparationfor the experiments we proceed as follows. First atubular section of a plasmodium is excised from alarge culture and implanted into the channel. Nextthe wells, i.e., the holes drilled in the circuit boardare filled with material from the anterior, fan-likeparts of the culture. By covering the top side ofthe circuit board with a thin, gas-permeable PDMSmembrane the plasmodium is fully enclosed, whichenables experiments of several hours duration. Aperspex frame holds the layers together. We referto this stack of agar sheet, laminate, (copper layer)PCB and PDMS as a “chip”. After an incubationperiod of 2–3 h in the dark, during which the mate-rial in the dumbbell shaped mask fuses into a singleplasmodial cell, the chip is ready for experiments.We typically prepare several plasmodia in chips andselect those that show a strong oscillation signal.

A custom circuit based on an impedance con-verter/network analyser (AD5933, Analog Devices,www.analog.com) and two multiplexers (ADG732,Analog Devices) in combination with custom soft-ware enables the monitoring of the cellular oscilla-tions through either a universal serial bus or I2Cinterface [22]. The actual implementation of thecircuit used in the experiments described here, is aprototype with easy access to all components andthe ability to isolate blocks for testing. We expectto reduce the layout to less than a quarter of itscurrent size (cf. Fig. 4). Impedance is measured at100 kHz applying an excitation of ≈1 V peak-to-peak.

The impedance circuit is connected through I2Cto a gumstix computer (www.gumstix.com), whichsamples all eight electrode pairs at ≈1 Hz andstores the information in flash memory. The datafrom all electrodes on both plasmodia on the chip

6

is recorded for later analysis. For driving the robot,however, one of the plasmodia is selected. Foreach of the two wells filled by this plasmodium,only the data from the pair of electrodes that showthe stronger oscillation is selected for further pro-cessing. To reduce noise the moving average over15 samples is computed (curves in Fig. 3). Thisprovides a sufficient signal even though not muchprovision for screening against noise was made inthe prototype, except separating the power sup-ply for the impedance circuitry from the supplyof the robot and gumstix. After the noise filter-ing, the drifting-component in the signal is removedby subtracting the 15 s delayed signal. Then thephase-relationship between the two signals origi-nating from the left and right well is classified ac-cording to whether both signals have equal or op-posite sign (bar at the bottom of Fig. 3).

The impedance circuit board carrying thePhysarum chip and the gumstix are mounted oncustom designed wheeled robot base [23]. A sim-ple rule is used to map the phase-relationship be-tween the two wells onto the motion of the robot:If the signs of the signals are equal the robot takesa random turn, if the signs differ the robot movesstraight forward. Within the 1 second period be-tween state updates the robot can move at most9 cm. The robot is confined to a round table with1 m diameter by means of an infrared sensor. If thesensor moves off the table, further “forward” com-mands are ignored by the low-level driver circuitof the robot until a rotation brings this constraintsensor back onto the table surface. For track-ing of the robot a network camera (AXIS 206M,www.axis.com) is mounted above the table and im-ages captured at a rate of 3 frames per second. Anilluminated arrow-shaped target mounted on therobot is located automatically in the frames to ob-tain the trajectory of the robot.

Acknowledgements

This research was funded in part by the Universityof Southampton Life Sciences Interface Initiative,

supporting FDR and ST. Funds from Microsoft Re-search Cambridge, The Nuffield Foundation andEPSRC (through the Nano Forum) provided sum-mer internships for JG, PM, and RS.

References

[1] R. Pfeiffer and C. Scheier. Understand-ing Intelligence. The MIT Press, Cam-bridge, MA, 1999.

[2] R. A. Brooks. The relationshipbetween matter and life. Nature,409:409–411, 2001.

[3] M. Conrad. The price of programma-bility. In R. Herken, editor, The Uni-versal Turing Machine: A Fifty YearSurvey, pages 285–307. Oxford Univer-sity Press, New York, 1988.

[4] M. Conrad. Scaling of efficiency inprogrammable and non-programmablesystems. BioSystems, 35:161–166,1995.

[5] S. Tsuda, K.-P. Zauner, and Y.-P.Gunji. Computing substrates and life.In S. Artmann and P. Dittrich, editors,Explorations in the Complexity of Pos-sible Life: Abstracting and Synthesiz-ing the Principles of Living Systems,Proceedings of the 7th German Work-shop on Artificial Life, pages 39–49,Jena, Germany, July 26-28 2006. IOSPress.

[6] K.-P. Zauner. Molecular informationtechnology. Critical Reviews in SolidState and Material Sciences, 30(1):33–69, 2005.

7

[7] S. Tsuda, K.-P. Zauner, and Y.-P.Gunji. Robot control with biologicalcells. BioSystems, 87:215–223, 2007.

[8] A. C. H. Durham and E. B. Ridgway.Control of chemotaxis in physarumpolycephalum. The Journal of Cell Bi-ology, 69:218–223, 1976.

[9] Z. Hejnowicz and K. E. Wohlfarth-Bottermann. Propagated waves in-duced by gradients of physiological fac-tors within plasmodia of Physarumpolycephalum. Planta, 150:144–152,1980.

[10] S. Tsuda, K.-P. Zauner, and Y-P.Gunji. Robot control: From silicon cir-cuitry to cells. In A. J.Ijspeert, T. Ma-suzawa, and S. Kusumoto, editors, Bi-ologically Inspired Approaches to Ad-vanced Information Technology (BioA-DIT’06), Osaka, Japan, January 26-27, 2006, Proceedings, volume 3853of LNCS, pages 20–32, Berlin, 2006.Springer.

[11] A. Takamatsu and T. Fujii. Time delayeffect in a living coupled oscillator sys-tem with the plasmodium of physarumpolycephalum. Physical Review Let-ters, 85:2026–2029, 2000.

[12] R. Halvorsrud, I. Giaever, and M. M.Laane. Patterns of oscillation dur-ing mitosis in plasmodia of Physarumpolycephalum. Protoplasma, 188:12–21,1995.

[13] F. D. Revilla, H. Morgan, and K.-P. Zauner. Impedance spectroscopyto measure contractile oscillations

in physarum polycephalum. Bio-Dielectrics: Theories, Mechanisms andApplications; Poster abstract., April2006.

[14] V. Braitenberg. Experiments in Syn-thetic Psychology. MIT Press, Cam-bridge, MA, 1984.

[15] J. Adler and W.-W. Tso. “Decision”-making in bacteria: Chemotactic re-sponse of Escherichia coli to conflictingstimuli. Science, 184:1292–1294, 1974.

[16] B. E. Scharf, K. A. Fahrner, L. Turner,and H. C. Berg. Control of directionof flagellar rotation in bacterial chemo-taxis. Proc. Natl. Acad. Sci. USA,95:201–206, 1998.

[17] W. J. Blake and F. J. Isaacs. Syntheticbiology evolves. Trends in Biotechnol-ogy, 22(7):321–324, 2004.

[18] D. E. Nivens, T. E. McKnight, S. A.Moser, S. J. Osbourn, M. L. Simp-son, and G. S. Saylery. Bioluminescentbioreporter integrated circuits: poten-tially small, rugged and inexpensivewhole-cell biosensors for remote en-vironmental monitoring. Journal ofApplied Microbiology, 96:33–46, 2004.doi:10.1046/j.1365-2672.2003.02114.x.

[19] R. Vijayaraghavan, S. K. Islam,M. Zhang, S. Ripp, S. Caylor, N. D.Bull, S. Moser, S. C. Terry, B. J.Blalock, and G.S. Sayler. A biore-porter bioluminescent integrated cir-cuit for very low-level chemical sensingin both gas and liquid environments.Sensors and Actuators B, 123:922–928,2007.

8

[20] Y. Tanaka, K. Sato, T. Shimizu,M. Yamato, T. Okanobd, and T. Ki-tamori. A micro-spherical heart pumppowered by cultured cardiomyocytes.Lab on a Chip, 7:207–212, 2007.

[21] M. J. Kim and K. S. Breuer. Con-trolled mixing in microfluidic systemsusing bacterial chemotaxis. Anal.Chem., 79:955–959, 2007.

[22] P. Macey. Imepdance spectroscopybased interfacing with a living cellfor biosensors and bio-coporcessors.Part III Project Report, School ofElectronics and Computer Science,University of Southampton, May 2007.

[23] G. Jones. Robotic platform for molec-ular controlled robots. Part IIIProject Report, School of Electronicsand Computer Science, University ofSouthampton, May 2006.

9