LARDE is a data extraction framework that intelligently reduces and persistently stores world model data for effortless future retrieval.
Autonomous systems continue to be outfitted with increasing numbers of sensors that are capable of collecting extremely large amounts of data over the course of a mission. Even autonomous systems with high storage capacities can run into storage limitations when burdened with large amounts of raw sensor data over long missions. Current data collection methods typically involve the storage of raw (sometimes compressed) or near-raw sensor data, such as images from video streams or 3D point clouds from LIDAR sensors or stereo camera systems.
Robotic Research, LLC teamed with Southwest Research Institute (SwRI) on this Phase II program to develop a Learning-based Approach for Relevant Data Extraction (LARDE). LARDE is a data extraction and handling framework that can intelligently reduce the volume of raw data from on-board sensors, and organize and persistently store the reduced relevant dataset. It is generic enough to be useful for a variety of current and future autonomous systems, but specific enough to directly support missions fielded with autonomous systems in the near-term. The LARDE framework stores compressed sensor and world model data in a persistent database. It contains the tools for inserting data into and querying data from the persistent data store for use in autonomous vehicle applications. As disk space reaches maximum capacity, the LARDE framework intelligently reduces the size of the persistently stored data.