Prev-Talk | Next-Talk | All-Talks | Talks-Sorted | MOOS-DAWG'19

Talk-21: Marine Perception Datasets: Sensor Evaluation, Interface, and Public Release

Paul Robinette, Michael DeFilippo, Michael Sacarny, MIT Sea Grant

In 2018 the AUV Lab started evaluation on the optimal marine sensor suite needed to safely navigate a marine vessel autonomously. In doing so we have started to create and release publicly available marine sensor datasets. We collected this data using our autonomous surface vessel the Remote Explorer 4 (REx 4), which was outfitted with a suite of sensors we believe is required for safe navigation of inland waterways. To accomplish this task we designed a concept perception box, with the intent to record and process data from each sensor. We will review the initial design of the perception box, as well as highlight its limitations, and subsequent upgrade. We will discuss our results with regards to the sensors evaluated in 2018, as well as how we plan to proceed in evaluation of new sensors to address the gaps in the previous sensor suite. Specifically, we have to address increased sensor load from the added sensors by developing a completely new perception box in 2019. Finally we will have a released set of sensor data for researchers to access of the following locations Lower Basin and Poor Manís landing, located within the Charles River in Boston, MA. Along with the data sets we have prepared for the release of our driver code for the sensors interfaced with MOOS-IvP in this research.

Categories:

  • USVs, Sensor Processing, Public Code/Data