Fronts have been recognized as hotspots of intense biological activity and are important targets for observation to understand coastal ecology and transport in a changing ocean. With high spatial and temporal variability, detection and event response for frontal zones is challenging for robotic platforms like autonomous underwater vehicles (AUVs). These vehicles have shown their versatility and cost-effectiveness in using automated approaches to detect a range of features. Targeting them for in-situ observation and sampling capabilities for frontal zones then provides an important tool for characterizing rapid and episodic changes. We introduce a novel momentum-based front detection (MBFD) algorithm which utilizes a Kalman filter and a momentum accumulator function to identify significant temperature gradients associated with upwelling fronts. MBFD is designed to work at a number of levels including onboard an AUV, on-shore with a sparse real-time data stream and post-experiment on a full resolution data set gathered by a vehicle. Such a multi-layered approach plays an important role in mixed human-robot decision making for oceanographers making coordinated sampling and asset allocation strategies in large multi-robot field experiments in the coastal ocean.

The Overbot is one of the original DARPA Grand Challenge vehicles now being used as a platform for autonomous vehicle research. The vehicle, equipped with a complete actuator and sensor suite, provides for an extremely capable robotic platform with computing infrastructure and software framework already in place to create a reconfigurable testbed. For point to point navigation, calculating suitable paths is computationally difficult. Maneuvering an autonomous vehicle safely around obstacles is essential, and the ability to generate safe paths in a real time environment is crucial for vehicle viability. We previously presented a method for developing feasible paths through complicated environments using a baseline smooth path based on cubic splines. This method is able to iteratively refine the path to more directly compute a feasible path and thus find an efficient, collision free path in real time through an unstructured environment. This method, when implemented in a receding horizon fashion, becomes the basis for high level control. In this work we perform Monte Carlo simulations to validate algorithm performance. The algorithm demonstrates a high success rate for all but the toughest of environments.