Siphonophora (2009)

Devices floating in a lake respond to aquatic life and conditions

Borrowing its name from a class of marine invertebrates, Siphonophora consists of a collection of small reactive devices that float in one of Gunpowder Park's ponds, tracking light, temperature, audio, pH levels and other pond life activity.

Through this project, a collaboration with Robert Davis, Psychology Department, Goldsmiths College, we researched ways to enable the population of devices to evolve their behaviour organically in response to specific site conditions including the local bird population, insects in and around the water and human interaction from visitors who come to see it from the existing bird hide.

Siphonophora's habitat, 2008

Situated at a bird hide to the edge of the lake was a pair of headphone sockets driven by amplified signals from hydrophones immersed at the lake edge. Here visitors could listen to the combined sounds of the lake water, bubbles rising from lake plants, the motions of aquatic insect life and the sounds from the cellular devices floated in the lake water.

Siphonophora's habitat, 2008, several weeks later

Also positioned at the bird hide was a hand cranked generator that supplied energy to an oscillator feeding an underwater speaker, the frequency of which was similar to the sounds emitted by the cellular entities, this provided a way for visitors to signal, or stimulate the floating components.

Siphonophora's habitat system diagram

Using Pachube an online data feed from sensor and positional information from the floating platform was arranged. The sensor package consisted of power supply, sensor circuitry, energy storage, and interface circuitry to a modem. This allowed sensor data to be transmitted in the form of text messages to an SMS gateway and rendered on the Pachube web site. Measurements were taken every 20 minutes with transmissions occurring every 4 hours. The sensor data consisted of 5 instrument channels and positional information as detailed below.

  • Accelerometer data for tilt/motion in 2 axes at right angles, orthogonal to the lake water surface
  • Temperature sensor, the temperature within the pachubox housing in shade, beneath the solar panel
  • pH probe data indicating lake acidity/alkalinity
  • Ambient light sensor mounted beneath the solar panel
  • GPS data positioning the platform as it moves around the lake area Magnetic compass data indicating the current bearing

Energy Storage was via 3.5Ah maintenance free sealed gel lead acid battery under cyclic charge from a 14.5V shunt regulator fed from a 4.5W polycrystalline solar panel. The charging circuit also included a low leakage diode to prevent losses into the solar panel at night.

View from Siphonophora

The sensor package enclosure was constructed from reclaimed acrylic sheet approximately 13mm thick, this had then been solvent welded watertight and any unwelded joints filled with silicone sealant. Nylon bolts were used throughout in order to account for thermal expansion of the plastic and to resist corrosion. In all places throughout the design an attempt has been made to preserve the transparency of the structure, this has the effect of lessening its visual impact as a collection of technological artifices and to imply a more biological presence, becoming almost cellular, having a transparent cell membrane enclosing a collection of functional blocks, much like a cell nucleus. It also enabled the structure to visually coalesce with the water surrounding it making the structure visually ambiguous and slight.

Siphonophora entities

The cellular entities were of two types, sensory cells and motor cells. The sensor cells served to monitor environmental conditions of motion/vibration, temperature and illumination and convert them into sound, temperature controlling the rate at which the tone pulsed, illumination controlling the pitch of the emitted tone, and motion the modulation and intonation of the tone burst.

The motor entities propelled the cell array in response to acoustic signals, they were also able to sense, using pulses of red light, objects and boundaries that they might collide with and avoid them. This is a reflex action inbuilt by design. Not all of the responses of the motor cells are fixed, but able to adapt over time in a self determined way.

Siphonophora entities

To avoid the use of batteries and toxic electrolytes super capacitors where chosen as the main energy storage device. Charge regulation was by shunt regulator using a light emitting diode and a string of signal diodes. During periods of high sunlight, when the battery or capacitor might be at full charge the shunt regulator dissipates the excess energy converting it into heat. It effectively shorts out the photovoltaic to reduce the voltage across it. As it does not switch the output of the photovoltaic it does not introduce any loss that might be associated with a transistor switching scheme. During periods of high sunlight and low charge the regulator lies dormant until the voltage across the energy storage device exceeds a critical value, then its resistance drops. By placing this circuit behind a low leakage diode, with respect to the charge storage device, any possible currents back from the charge storage device to the solar panel were minimised.

The cellular devices were constructed from two symmetric halves that have been vacuum formed on a purpose built vacuum forming bed and solvent welded together to form a lenticular shell. The material itself being PETG, a thermoplastic that vacuum forms easily and can conform to tight detail, approved for contact with food it offers no particular toxicity to the aquatic environment. It is also optically clear, again so that photovoltaic power might be harvested and so that it might appear more in keeping with a biological entity than a collection of technological artefacts. Each cell was hermetically sealed so that signals to and from each cell must be transmitted as light, motion or sound through the cell membrane. Each cell has a volume of around 750cc, approximately half of which is filled with circuitry and an optically clear silicone encapsulant. The encapsulation further protects the circuitry from water and also lowers the centre of gravity of the cell so that it floats half submerged as the encapsulant has a density of approximately that of water. It also allows sound to be transmitted efficiently to and from the cell and to some degree, owing to its thermal inertia, to stabilise the temperature of the circuitry.

Siphonophora crank at bird watching hide

The behaviour of the sensory cells is fixed in the way that it responds to environmental parameters, while the behaviour of the motor cells is not. The motor cell behaviour is free to adapt to certain environmental parameters with two caveats.

First, that the cell is primarily phototropic when the energy storage is depleted. That is that it has a propensity to close distance with light sources. As a photovore the cell would have evolved this type of behaviour via selection as without it the cell would be unable to persist in its other behaviours – effectively electing to ‘starve’ during periods where light was short or a different behaviour than ‘feeding’ was favoured.

Second, That it will move aside when there is a perceived boundary ahead. The motor cell has innate collision avoidance behaviour. To avoid the platform becoming beached this behaviour has some precedence above other behaviours that the cell might exhibit.

The remainder of the motor cell behaviour is centred about phonotropism, that is the tendency to track and follow sounds. All of the above behaviours also included a degree of down regulation during periods of over excitement, which is that when over stimulated in a particular way the cells behaviour changes to regulate out the source of this stimulus, becoming exhausted.

The cells are excited by novel events when other behaviours are not so strong as to override them. The circuitry that determines the behaviour of the motor cells is analogue in design and is an implementation of a spiking neural network, a parallel to the processing that occurs in the nervous material of biological entities. There is no digital processing, no conventional computer or microcontroller involved in determining the behaviour of the motor cells, all processing is parallelistic and continuously valued in either time or intensity.

This affords the neural circuitry a more biologically plausible mode of operation and behaviour. The neural network approach to the design of the circuitry allows the device to exhibit behaviours that might not be predicted or indeed predictable by the designer, this allows greater freedom for the design to find its own tolerances, tendencies and habits, rather than the designer to impart these by deliberate action. This is the core idea that allows the cellular platform to evolve its own behaviour patterns. As the components used are all of slightly different values, as a result of the manufacturing process, individual differences are inherent in the construction and the behaviour of each of the cells, this is true of both the sensory cells and the motor cells as they are both of analogue construction. If the dynamics of an analogue information processing system are even the most trivial of functionality, there will always be small differences between what appear to be the same circuit by design as the components used will all vary a little from there stated value. Most resistors and capacitors, the most basic of passive electronic components are commonly of plus or minus 5% of their stated value, so too varies the performance with temperature and the forward voltage of a simple semiconductor diode, or the gain of a transistor, making these systems quite sensitive to their environment. If a system contributes some data back into the environment from which that data were obtained then there is the distinct possibility that the system might develop strongly nonlinear dynamics.

Research began in August 2008 and Siphonophora lived in the park from January to March 2009.


Original project outline

We propose to build structures and devices in and around the Osier Marsh area of Gunpowder Park, some of which are accessible to people but some of which are not. The main components will consist of a floating jellyfish-like structure in the largest water body and additional interventions at the bird watching screen. We are interested here in looking at 'tropisms' and 'mimicry' and how to create something that grows and evolves along with the ecology that it is growing in.

There will be two main components to the installation.

The first is a floating "jellyfish" platform in the lake, approximately 1.5m x 1.5m. Over time we will add more and more components to this device and ultimately it will be able to propel itself and interact with wildlife and visitors to the site.

This structure will exhibit various tropisms -- that is, a tendency to be drawn towards certain types of event and energy. This is somewhat similar to animals in that it allows it to feed efficiently on various sources of energy. In this, it may start to mimic the behaviour of various birds in the lake and occasionally to attract people who might be observing from the bird watching screen area.

A well-known example of a robot that exhibits tropism is Grey Walter's turtle that followed light sources. Here our robotic raft might initially be attracted to water movement (generated by birds and possibly large fish or people kicking the water's edge). When it detects movement from a particular direction it will head in that direction -- the aim is to capture some of the water movement that it has to detected to generate more energy, which will enable it to move more in the future. Of course, if it expends more energy in doing this than it gains then it will adapt itself to minimise waste in the future and might instead become light-tropic - moving towards areas of brighter light, where it can recharge its solar cells. The "jellyfish" might even harvest energy from the small variations in temperature found across the lake. The idea is for it to learn how to sustain itself - it might well discover that provoking fish to splash is the best way to recoup enough energy to continue surviving. Or it might learn that moving itself towards people standing at the bird hide is better, because they will be invited to contribute energy in other ways.

Analogue neural networks from our previous project, the Evolving Sonic Environment, will be the basis of the behaviour determining component and will allow the behaviour to become adaptive, enabling the "jellyfish" to decide between which sensors might indicate the path to viable sources of energy and interaction.

Initially a floating platform tethered in the middle of the lake and unable to move very far, this system will support various sensors and data logging/SMS messaging hardware. It will most likely be constructed from a number of intelligent cellular components, each hermetically sealed and waterproof, loosely coupled together to form one ‘colony’, only able to move under the influence of ripples. Most of the structure will float just under the surface of the water and will be only partially visible for most of the time. Where possible recycled and environmentally sustainable materials will be used, such as cork and balsa for floats.

The second component is an intervention at the bird-watching hide, where people view the lake. Here, a number of sensors and actuators will respond to people observing the wildlife and might allow people to interact with the floating component and to listen to various metabolic processes of the floating entity and of the surrounding environment to which it is sensitive. Visitors might be able to signal their presence to the floating platform and encourage it towards them. A crank operated generator will be secured to the steel work of the bird watching screen. This will allow visitors to input "energy" into the lake in order to ‘feed’ and attract the waterborne component of the installation.

The installation will be incremental, as various parts of the system are tested and the relationships between various signals found.

As a starting point, we will simply count the number of times that the crank has been operated -- when left with no instructions -- and measure how much energy might be harvested.

Water ripple detection will be used to determine if useful amounts of energy might be harvested from the lake, and will provide a way for the floating structure to detect animal and human actions. Both uni- and omni- directional sensors will be prototyped as might also an energy harvesting ripple device using electromagnetic or pneumatic methods.

Temperature sensors will be placed onto the sensor platform so that we might examine the dynamics of air/water temperature for energy harvesting and sensing of power sources, ie thermotropism. This will consist of both silicon integrated temperature sensors and a Peltier heat pump used to generate current from difference in temperature between lake water and the air via a pair of heat sinks.

Light sensors will be used to see how much energy might be collected via solar panels and might be used to detect the motion of creatures near the floating platform in order to mimic in some way the behaviour of these creatures. Light sensing will also be used to generate phototropism in the later stages of development of the floating platform.

The interaction and dynamics of all of the sensor data will be examined for any interesting features that might become special events in the behaviour of the system.