Compressive Robotic Systems: Gaining Efficiency Through Sparsity in Dynamic Environments

Sponsor: National Science Foundation

Award Number: CMMI-1562031

Abstract:

This project investigates autonomous control and coordination of a group of robots that are tasked to explore, map, or monitor the environment they are in. The project aims to enhance the capabilities of such a group of robots by integrating Compressive Sensing for data compression. Compressive sensing enables robots to quickly extract information from their environment, efficiently communicate that information to each other over a wireless network, and intelligently direct their motion to obtain relevant sensing data in the future. Significant theoretical and technical challenges must be addressed in this project to realize the potential of a compressive robotic sensing system. The project will demonstrate results in two specific applications, (i) driving a group of aerial robots to monitor their environment, (ii) driving robotic micro-probes to measure processes inside a living cell. The project also seeks to disseminate its findings through educational and outreach activities. Results will be incorporated into undergraduate and graduate level courses in control theory at both Boston University and Stanford University. The researchers will also work with high school students and undergraduates through research mentorship programs and through lab demonstrations for visitors.

The fundamental goal of the project is to create rigorously analyzed algorithms that take advantage of sparse signal descriptions to create efficient motion plans for a team of sensing robots that monitor the environment. The driving hypothesis is that sparsity can greatly extend the performance of robotic sensing systems by saving battery power, computation, storage, and communication bandwidth—all critically limited resources for robotic platforms. The research team will take a Bayesian approach to Compressive Sensing, which allows for sensing quality to be quantified with information theoretic metrics such as entropy. A receding horizon control approach will be developed for driving robotic sensors to collect the most valuable sensor data, in order to reconstruct a sparse representation of their environment using Compressive Sensing. Such control strategies will be adapted to both static and dynamic environments, and both centralized and distributed solutions will be sought. The concepts developed in this project will be applied to two specific sensing domains: (i) networks of quadrotor sensing robots sensing environmental data and (ii) confocal fluorescence microscopy for three-dimensional imaging of dynamics in bio-molecular systems. These two application domains have radically different length and time scales, dynamical properties, and information content. A successful application of the ideas developed in this project to both these domains will prove the generality of the Compressive Robotic Sensing System concept.

For more information: click here.