Events Calendar

PhD Defense - Jnaneshwar Das

Title: Data-driven Robotic Sampling for Marine Ecosystem Monitoring

Candidate: Jnaneshwar Das
Committee members:
Gaurav S. Sukhatme (chair)
Stefan Schaal
David Caron (outside member)

Time: December 19, 2013, 12 noon
Place: Ronald Tutor Hall (RTH), room 406

Abstract:

Monitoring marine ecosystems is a challenging problem due to complex ocean dynamics and limitations of in-situ sensing. Sensors onboard Autonomous Underwater Vehicles (AUVs) measure environmental and optical properties such as temperature, salinity, and chlorophyll fluorescence in real time. However, studying plankton ecology and community structure requires retrieval of water samples to shore for lab analysis using molecular or morphological methods. This necessitates manual sampling from shipboards or piers -- a resource and time intensive process that makes persistent monitoring of marine ecosystems difficult.

Motivated by advances in AUV technology that allow autonomous retrieval of water samples, we have developed a data-driven robotic sampling paradigm for ecosystem monitoring. We treat microorganism abundance as a hidden feature that can be predicted using a probabilistic regression model trained on past data, with observable environmental parameters as its input. This model is used onboard the AUV to predict microorganism abundance in real time. An online sampling policy uses these predictions to make irrevocable decisions to sequentially collect a fixed number of water samples over a course of a deployment. Learning a probabilistic model facilitates introspection, both online, guiding sampling decisions in an exploration-exploitation setting, and offline, enabling scientists to be aware of the accuracy of the environmental niche of the organisms they are interested in, as captured implicitly in the trained model.

The contributions of this thesis are as follows. First, we explore a challenging science problem in a statistical machine learning setting, enabling the utilization of theoretically sound methods from function approximation, active learning, and online algorithms. Second, we present extensive studies on real field data from a week long campaign in 2005 consisting of 17 consecutive AUV deployments. The empirical evidence affirms the utility of our sampling approach. Third, and most important, we present results from a recent field deployment that targeted a genus of phytoplankton known to cause potentially toxic blooms. A probabilistic regression model was trained on a dataset of lab analyzed water samples from AUV missions carried out during a previous season. This trained model was used on board the AUV to target the phytoplankton of interest, and samples were analyzed on shore. Preliminary lab analysis results are promising, showing abundance of the target organism, and facilitating lab cultures. This is the first time such a field experiment has been carried out in its entirety in a data-driven fashion, in effect 'closing the loop' on a significant and relevant ecosystem monitoring problem.

Although the experimental context for this thesis is marine ecosystem monitoring, our work is well-suited for autonomous and persistent robotic observation of any hidden feature that cannot be measured in-situ, but possesses observable covariates, e.g for soil sample analysis, and surficial geology applications thus opening up the potential for advanced autonomous robotic exploration of unstructured environments that are inaccessible to humans.

Location: Ronald Tutor Hall of Engineering (RTH) - 406

Audiences: Everyone Is Invited

Contact: Lizsl De Leon