OVERVIEW

A variety of naturally-occurring and introduced microorganisms adversely impact marine ecosystems and uses of marine resources. They can affect human health, fisheries and tourism. However, conditions under which aquatic microorganisms develop are not well understood, and methods for detecting microorganisms are too slow for timely intervention. In-situ sensor networks provide one way to monitor microorganisms in real time. The goal of this project is to deploy large numbers of sensors and robots operating in a semi-autonomous but coordinated fashion in aquatic environments. The system should be able to identify and study the behavior of microorganisms in situ and in real time. The robots should have the ability to sample the aquatic environment with fine spatial and temporal resolution. Given the size of the marine environment, it is not likely that we can deploy the sensors with such a high density that they can provide the high spatial resolution we need. However, since the robots have the ability to move, they can redeploy themselves, and increase their density where relevant, providing a capacity for relatively few sensors to sample the environment with fine resolution. This project straddles the M³ and the Sensor-Coordinated Actuation areas.

APPROACH

Thermocline monitoring using a wireless underwater network

We have designed and constructed an underwater sensor/actuator network to detect temperature gradients. A region of sharp temperature change (thermocline) is an important physical boundary affecting the proliferation of marine microorganisms. We have developed a distributed algorithm using local communication based on binary search to locate a thermocline within a water column using a mobile sensor network. Experiments using our mote-based test bed demonstrate the validity of this approach. We have also recently performed experiments with a submarine robot as a data mule to improve the energy efficiency of the network. Comparison between experimental data with and without the data mule shows that there are considerable cost savings in the sensor network due to the data mule.

The basic idea of our approach is sampling by divide and conquer. We deploy sensor nodes in a vertical array such that communication range is limited and each node can only communicate with its neighbors. Each node is able to acquire temperature data. We assume each node is localized (this is not unreasonable since the nodes are attached rigidly to a rod, which moves vertically causing all nodes to move together). The search space is 1D, and is divided into regions. Every node uses its ability to move to explore one of those regions. The process is refined by splitting regions into halves, i.e., binary search. Each node communicates with its peers and tries to persuade them that the thermocline lies within its search space. A process of data aggregation is enacted on the route from each node to the user to combine the conclusions (about the thermocline location) arrived at by the various nodes.

Bacterial Navigation & Applications to Marine Surface Sensing

Locating gradient sources and tracking them over time has important applications for environmental monitoring. Our approach is inspired by bacterial chemotaxis. Robots navigate to sources using gradient measurements and a simple actuation strategy (biasing a random walk). Extensive simulations show the efficacy of the approach in varied conditions including multiple sources, dissipative sources, and noisy sensors and actuators. We have also demonstrated how such an approach could be used for boundary finding. We have validated our approach by testing it with a small robot (the robomote) in a phototaxis experiment. A comparison of our approach with gradient descent shows that while gradient descent is faster, our approach is better suited for boundary coverage, and performs better in the presence of multiple and dissipative sources.

The ability to autonomously detect, locate and track such phenomena (the source of the induced gradient) would give scientists a tool to monitor and study ecosystems at an unprecedented level of detail. We are particularly motivated by the research goal to track changes in the spatial distribution of blooms of A. anophagefferens in nature. This information can be correlated with measured parameters that might govern the abundance and distribution of the alga, including temperature, nutrient concentrations etc.

Bacteria sense chemical concentration using receptors. They are able to detect temporal and spatial changes in chemical concentration based on the fraction of receptors occupied at successive time intervals. Bacterial motion alternates between two stages (run and tumble). The duration of the run (which is related to the mean free path) is dependent on the concentration gradient that is sensed in the vicinity of the bacterial cell. In the absence of a gradient, the run length is independent of the direction of motion and the bacterium executes a random walk. In the presence of a positive gradient (increasing concentration of a chemoattractant), the frequency of tumbling is reduced resulting in a longer run length. A negative gradient does not have any effect on the tumbling. This change of tumbling frequency in response to concentration gradient results in chemotaxis, allowing bacteria to move preferentially towards nutrient sources.

Figure 1- Difference between biased vs. an unbiased random walk

We are interested in the development of simple, robust, energy efficient and cost-effective techniques which could be used in-situ to locate source phenomena of interest to scientists. We propose a simple strategy for a mobile robot (or multiple robots) to navigate to such a source using gradient information and extremely rudimentary actuation. Our strategy is inspired by the studies of taxis in bacteria.

The strategy of such a bacteria-like robot can be summarized as “sense and move”. A robotic node executing a biased random walk has very small requirements in terms of memory since only the last sensor reading needs to be stored. The processing requirements are minimal since the only processing required is comparison between successive sensor readings (gradient computation). Only a minimal amount of motion control is required to hold the heading of the robot in a particular direction for a particular duration of time (depending on bias levels).

SYSTEMS / EXPERIMENTS

Thermocline monitoring using a wireless underwater network

Figure 2– (left) a schematic of the sensor network used, and (right) a photograph of the experimental testbed.

A heater was used to create a thermocline in the tank. This heater was placed in the water just beneath the surface. In the steady state, a thermocline is established at depths between 200mm and 400mm where the temperature drops rapidly. A series of experiments were carried out to locate this thermocline. These are shown in Figure 3. The four figures with solid lines demonstrate the estimation of thermocline location and the four figures with stars indicate actual temperature readings as a function of depth.

Figure 3- Adaptive sampling using binary search - experimental results

After the first one or two steps of the distributed binary search, most sensor nodes become inactive. However, they have to be awake if they are on the path from the active nodes to the root of the routing tree since they are needed to forward messages from the active node to the users. One way to create a short cut and thereby save energy under water is to use a messenger, a robotic node that can move itself autonomously. This submarine robot would move from the neighborhood of the root to the active node, and forward the messages from one to another, thus acting as a data mule. Motion consumes energy; but we assume that a process exists to recharge the robot when it surfaces, and do not analyze it further here. Our experiments show that the introduction of such a data mule reduces energy consumption of the static network. We give a brief description of the robot below. Further design details as well as results from datamule experiments are discussed in the Actuation section.

The robot is based on the mote platform, which is used extensively in the sensor networking community as an experimental test bed. The physical system is shown below in Figure 4. The robot is composed of two parts: the base on which all the electronics are mounted, and the housing which is a protective enclosure. The robot is a cylinder standing 23.5 cm high and 6 cm in diameter. So far the tests conducted prove the feasibility of the platform. We are poised to conduct further experiments in underwater autonomous sensor networks. The tests with the pressure measurements vs. depth were shown to be linear and are very accurate for our purposes. The robot is able to regulate its depth within 5 cm of the desired depth and consequently we are able to obtain a fairly accurate plot of the thermocline region in our tank. We are in the process of designing a more accurate depth regulation system (+/- 2.5 cm).

Figure 4- (left) the submarine robot and (right) the robot making a dive in the tank

Bacterial Navigation & Applications to Marine Surface Sensing

Simulation Study: A customized simulation platform was designed for evaluating the implications of a bacterial motion-based approach for locating and tracking gradient sources and their boundaries. The simulations were modeled for a wide range of environments and gave insight into the nature of applications for which this class of algorithms is well suited. We studied the effects of different parameters on our algorithm. For example the effects of varying bias are shown in the figure below. Increasing the bias resulted in more rapid convergence. On the left in Figure 5 we show robot displacement vs. time. In the center, we show the distance between the robot and the source vs. time, and on the right we show the percentage of robots (in a multi-robot setting) at the source vs. time.

Figure 5- Effect of varying bias.

This technique (Figure 6) did not show preferential movement towards a particular source based on the time of appearance of the source. Simulations carried out with a number of sources of different intensities demonstrated that all sources were tracked, although the weaker sources received a comparatively smaller fraction of the robots where the gradients were small. However, on average, all sources were well covered.

Figure 6- The ability to efficiently respond to multiple sources

We performed another set of simulations introducing a Gaussian error in the decision function modeling the non-static sensor errors and actuation errors (i.e., motion of the robot might not be the same as the command signal applied). Even in the presence of 40% error, the robots still converge to the gradient source as can be seen in Figure 7.

Figure 7- The effect of errors

Another set of simulations were performed to understand how well the robots spread around a gradient source. Figure 8 presents the result of initializing all the robots at a single location at one corner of the grid. As can be seen, the robots approach the source from all directions. Similar results were obtained from other robot deployment strategies.

Figure 8- The effectiveness at boundary detection.

We also evaluated our biased random walk strategy against a gradient seeking strategy. The results of single-source seek experiment indicate that a simple gradient descent scheme performs better than our biased random walk for small bias values. However, as the bias levels are increased, the two become comparable.

The results from the simulation modeling multiple time-varying sources indicates that with the introduction of a second source, the gradient descent strategy keeps following the first source, thus the number of robots at the first source keeps increasing and very few robots reach the second source. On the other hand, the results from the biased random walk are quite impressive. Both gradient sources are simultaneously tracked and receive a fair share of the number of robots reaching them. The introduction of more gradient sources results in some robots tracking each one of them as long as there are some available robots. Thus for the purpose of tracking multiple gradient sources, our algorithm clearly outperforms the gradient descent approach.

Figure 9- Comparison with gradient descent strategy (single source)

Figure 10- Comparison with gradient descent strategy (multiple time-varying sources)

Experiments with the Robomote Platform: We chose the robomote as it provided us with a small, low-cost, platform with multiple mobile robots which could be easily programmed and used to test our algorithms. The robomote has been developed at CENS and (like the submarine testbed) is based on the mote. For details on the robomote testbed see the Actuation section of the report. We generated a light gradient using a light source placed at one end of the test-bed. A basic sensor board with a photo sensor was mounted on the Robomote (Figure 11) to sense the light gradient. The position of the robomote on the test bed was tracked using an overhead camera which captured frames and passed these to a tracker for data analysis and storage. Color blobs were mounted on top of the Robomote to help in the detection of Robomote location on the test bed.

Figure 11- Experiments with the robomote.

We used the two basic components move and rotate written in TinyOS for controlling the robomote to carry out the biased random walk. We positioned the robomote at a distance d (d=40cm, 80cm, 120cm) from the source. Note that by fixing d, the heading of the robomote still was a random variable and could be towards or away from the source. A small circle of 5cm around the source was considered as the source radius. We were interested in measuring if the robomote reaches the source, and if so in how much time? The speed of the robomote was set at 2cm/s with a turn time of approximately the same duration.

Once switched on at a distance d, the robomote starts off by taking a sample using the photo sensor. It moves along a straight line in the direction of its current heading for a distance and/or duration as specified in the random walk parameter MFP. At this point it takes another photo sensor reading and compares it with the previous reading from the photo sensor. If it senses no change or a negative change in gradient, it randomly chooses a new heading direction and rotates in place to orient to that direction. If a positive change in gradient was sensed, it continues its motion for an additional distance specified by its bias value before randomly computing the new direction and making a turn (’tumbling’). In either case the procedure is repeated by moving along a straight line in a particular direction for a distance and/or duration as specified in the random walk parameter MFP, followed by another decision based on the photo sensor reading. The experiment terminates when the robot reaches the source.

Each of the d values constituted a circular arc on the table of radius d units from the center of the light source. We repeated the experiment 75 times for each of the d values with random starting orientations and starting locations on the arc, and averaged our position readings between the gathered data for our analysis. The results from the robomote platform agree with the results we obtained from our simulation work.

Figure 12- Tracking (a)single photo-gradient source and (b)multiple photo-gradient sources.

As an application of the navigational algorithms described above, we have designed and constructed a 1m long air boat to use as an autonomous profiler and sampler in small bays off the California coast and Long Island, NY (the latter environment has recurring harmful algal blooms of our model organism, Aureococcus anophagefferens. The boat is equipped with a fluorometer that measures chlorophyll a content, a biologically meaningful indicator of plankton abundance. Other sensors measure temperature, salinity, and geographic position have also been incorporated to provide an environmental context for the biological data. The boat will autonomously profile a body of water along a pre-set pattern (using either GPS-based positioning), or using a bacterial chemotaxis-inspired algorithm. The latter method uses a biased random walk algorithm to migrate towards areas of high concentration of chlorophyll (using rapid fluorometric detection). An on-board, custom-designed water sampler allows for sample collection for later off-line analysis. We will incorporate an on-board biosensor specific to A. anophagefferens as this technology advances. For a description of the design of the boat see the Actuation section.

Test bed experiment: Stimulation of a ‘brown tide’ in a thermally stratified column

One of our main test beds for developing technology for adaptive environmental sampling, and for testing new methodology for marine microbial detection and enumeration, is a 2-meter glass column. We have established conditions under which A. anophagefferens can be grown in an ‘artificial water column’ in this glass column. Light is supplied to the column using a 250W metal halide plant grow bulb with remote ballast on a 12:12 light:dark cycle. The alga grows preferentially in the warm upper water and the thermistor strand and thermocline-finding algorithm (developed in the adaptive sampling project) are employed to direct the depths at which samples are taken (Figure 13). Within 2 weeks under nutrient-replete conditions the concentration of A. anophagefferens increased to 2 million cells per ml, and after 3 weeks the concentration was >5 million cells per ml (Figure 14). These abundances are consistent with true brown tides in natural lagoons of the Middle Atlantic coast (Long Island-Maryland) where natural brown tides occur.

Figure 13 - Temperature profile in a water-column in which thermal stratification has been induced.

Figure 14 – Increases in A. anophagefferens abundance during a time-course experiment.

Abundances of the harmful alga in this test bed were monitored effectively using off-line methods (ELISA). Algal abundances increased gradually over a two-week period, with higher abundances generally detected in the warmer upper waters, as expected. Once the A. anophagefferens population had attained high abundances of 5 million cells/ml, a low abundance of the predator, Pedinella hexacostata was added to the column. P. hexacostata is actually mixotrophic (conducts both heterotrophy and autotrophy), making it both a predator and (ultimately) a competitor of A. anophagefferens. The population of A. anophagefferens in the column in the presence of P. hexacostata was depleted significantly over the course of one week to less than half its concentration in the surface waters (Figure 15). Contrary to conditions before grazing began, algal cells were more abundant in the deeper, colder water during the grazing phase, an indication that the predator did not readily consume prey below the thermocline. Alternatively, this result may indicate the buildup of algal cellular debris below the thermocline that retained antigenic character.

Figure 15–BT concentration as a function of depth after additon of Pedinella. Data was taken 21 (T21 curve), 22 (T22), 24 (T2) and 27 (T27) days after the predator’s introduction.

Test bed experiment: Vertical migration of a red tide dinoflagellate in a stratified water column

We have also begun studying a toxic dinoflagellate, Lingulodinium polyedrum (Figure 16), using CENS approaches. A thermally stratified water column was used to study the vertical migration of L. polyedrum, blooms of which form red tides off the southern California coast with important ecological implications. L. polyedrum was maintained in culture and inoculated into the 2-meter high column of f/2 growth medium at a concentration of 1500 cells/mL. The column had a sharp thermocline at 110 cm depth. Light was supplied to the column using a 250W metal halide plant grow bulb with remote ballast on an 11:13 light:dark cycle.

Figure 16- Lingulodinium polyedrum (left) in high magnification (cells are approx. 30 µm), and a natural ‘red tide’ caused by the alga (right).

Over the course of three days, samples were extracted from depths throughout the column and the abundance of L. polyedrum was determined via direct microscopy and a newly developed quantitative polymerase chain reaction (QPCR) approach. We were thus able to track the vertical distributions of L. polyedrum in the column over 3 diurnal cycles in an effort to determine the nature of any vertical migration. Distinct differences were observed in the vertical distribution of cells over the course of a day; specifically, abundances were much greater in surface waters during ‘daylight’ hours while cells were more homogeneously dispersed throughout the upper portion of the water column during dark periods (Figure 17).

Figure 17- Vertical distribution of L. polyedrum in water column test bed. L. polyedrum was observed in high abundance in surface waters during ‘daytime’ hours as shown by the red discoloration in (A). L. polyedrum abundances were determined through direct microscopical counts (B, solid lines) and QPCR (B, ‘x’ with error bars), providing a direct comparison of the two methods.

The highest abundance of L. polyedrum occurred at the surface around noon while at ‘night’ there appeared to be a slight subsurface maximum at 100cm, just above the thermocline which was most pronounced around 110cm. Also of interest was the accumulation of cells at the surface in the ‘pre-dawn’ sampling at 05:00, suggesting the cells were able to anticipate the impending incursion of light. It thus appears that L. polyedrum does indeed adjust its vertical position in the column in response to light; the exact nature of this response will be studied further using the column test bed, changing light regimes, and grazing pressure.

Development of a sensor network for deployment in aquatic ecosystems

Major progress has been made during this work period on the design and construction of a sensor network (and incorporation of navigational control of our mobile sampling robot into the network). The network is comprised of 10 stationary nodes (buoys) that will be equipped initially for sensing temperature and chlorophyll fluorescence (a proxy for phytoplankton biomass) (Figure 18). These biologically-pertinent environmental characteristics will be collated into a 2/3-D picture of the ecosystem, and used to guide our small autonomous surface vehicle to desired sampling locations and collect and retrieve samples.

Although the immediate impact of this network will be modest (scientifically speaking), the proof-of-concept that it represents will be huge. The network will be fully wireless vis-à-vis transmission of data within the network and communication with and direction of the mobile sampling robot. The stationary nodes will collect environmental data and communicate that information throughout the network in real time. Information from the nodes will be analyzed within the network, and the mobile sampling robot will be directed to features of biological interest within and around the static nodes for high-resolution feature mapping and/or directed sample collection. Our plan is complete network construction by April 2005, and deploy the network and the robotic sensor/sampler in Lake Fulmor at the James Reserve in April/May 2005.

Figure 18- A stationary node within our wireless sensor network. For details, please see the Actuation section.

ACCOMPLISHMENTS This Year

• An improved version of distributed binary search for adaptive sampling was demonstrated to work on the underwater testbed. This testbed has been ‘on’ for large periods of time over the course of the previous year, and provides regular uninterrupted temperature measurements during experiments.
• Conversion of the wireless testbed to Mica2, and including energy-saving protocol S-MAC
• An experimental analysis of the bacterium-inspired navigation revealed that we can trade bias for speed vs. efficiency. Higher bias values accelerate the movement of the robots towards the source but might not be good if the source is mobile or has a variable intensity. Lower bias values result in longer response times for the robots to reach the source but are more effective in tracking weak, mobile and intensity-varying sources. Our results also indicate that it is possible to determine a range of bias values to track a gradient source with any desired source tracking characteristics.
• The source boundary detection results highlight the suitability of our approach to a set of applications where the source dimensions are comparatively large and we need to track its complete boundary (e.g. the boundary of an underwater plume or the boundary of an oil spill).
• A design and first build of a robot submarine for experiments in underwater sampling
• A design and first build of a 1m long air boat to use as an autonomous profiler and sampler in small bays off the California coast and Long Island, NY.
• Design and construction (nearing completion at this time) of a wireless sensor network for deployment into natural aquatic ecosystems.
• Performance of ‘bloom stimulation’ experiments with A. anophagefferens, using a thermally stratified water column and sensor-directed sampling.
• Performance of ‘vertical migration’ experiment with Lingulodinium polyhedrum using a thermally stratified water column and sensor-directed sampling.

FUTURE DIRECTIONS

A central theme in our application involving marine microorganism monitoring has been the development of sensing networks that can make numerous, low-cost observations within the environment, process that information locally, and guide more-expensive, less-densely-spaced biological sensors and/or samplers. To date, our work has been focused in the lab using our test beds (two large aquaria). This work has been instrumental in addressing various issues involving sensing, sampling and communication in aquatic environments as described above. We will continue to use our test beds as tools for the development and ground-truthing of new ideas and technological approaches for marine microorganism monitoring. Most importantly, we will move our work into natural ecosystems during the coming year. Specifically, we will:

• Test and verify our boundary detection and source-following strategies, in real environments, which we have so far studied in simulation and found to be effective.
• Improve and augment the capabilities of the existing robot boat to enable verification and validation of the bacterial navigation strategies, as well gather data in the field.
• Complete our wireless sensor network for deployment into natural aquatic ecosystems.
• Deployment of the network and robot into Lake Fulmor of the James Reserve in April-July. The JR deployment will allow full deployment and debugging of the static sensor network and the network-actuated mobile sampling robot, validation of the ability of the network to collect and collate environmental data, and the ability to direct the mobile sampling robot to features of interest.
• Deployment of the network and robot into Thompson’s Reservoir, a drinking water reservoir located on Santa Catalina island is being discussed with the appropriate parties at this time.
• Performance of experimental work in our water column test bed to examine the factors controlling the growth and behavior of harmful algal species.

PEOPLE

FACULTY

Prof. David Caron
Prof. Deborah Estrin
Prof. Aristides Requicha
Prof. Gaurav S. Sukhatme

STAFF

Carl Oberg

STUDENTS

Vitaly Bokser
Amit Dhariwal
Eric Shieh
Bin Zhang