MultiRobot Informed Path Planning Under Communication Constraints Brad

Multi-Robot Informed Path Planning Under Communication Constraints Brad Woosley and Raj Dasgupta C-MANTIC Lab Computer Science Department University of Nebraska at Omaha Presented at the UNO Research and Creative Activity Fair 2017 Brad Woosley – Research and Creative Activity Fair 2017 1

Introduction Exploration of unknown environment Sample and learn distribution of some phenomena Water, soil and rock samples, radiation, etc. Samples must be communicated back to base station Must be able to venture out of base station communication range for wide area search Battery constraints limit the time the robot can take and distance it can travel Brad Woosley – Research and Creative Activity Fair 2017 2

Related Work Frontier based exploration [2] Not directed to locations with high information gain Doesn’t handle communication constraints Build communication tree of robots [3] Limited range by number of robots Divide into explorer and relay robots [1] Allow breaking communication links Uses frontier based exploration Model phenomena of interest using Gaussian process [4] Doesn’t consider communication constraints [1] De Hoog, J. ; Cameron, S. & Visser, A. Autonomous multi-robot exploration in communication-limited environments Proceedings of the 11 th Conference Towards Autonomous Robotic Systems (TAROS), 2010. [2] Franchi, A. ; Freda, L. ; Oriolo, G. & Vendittelli, M. The Sensor-based Random Graph Method for Cooperative Robot Exploration IEEE/ASME Transactions on Mechatronics, 2009, 14, 163 -175. [3] Mukhija, P. ; Krishna, K. M. & Krishna, V. A two phase recursive tree propagation based multi-robotic exploration framework with fixed base station constraint IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2010, 4806 -4811. [4] Stranders, R. ; Rogers, A. & Jennings, N. A decentralized, on-line coordination mechanism for monitoring spatial phenomena with mobile sensors Intl. Joint Conference on Artificial Intelligence, (IJCAI) 2008. Brad Woosley – Research and Creative Activity Fair 2017 3

Gaussian Process Way to learn a function based on training data and samples taken As more data collected, better approximation reached Used for large scale data analysis Blue line predicted function Shaded area uncertainty +'s sampled points Image Source: http: //www. gaussianprocess. org/gpml/code/matlab/doc/ Brad Woosley – Research and Creative Activity Fair 2017 4

Proposed approach Two level approach Gaussian processes Information points Communication points Select best point from information and communication points Best overall utility Alternate between point types Brad Woosley – Research and Creative Activity Fair 2017 5

Information collection GP Input: 2 D - Point in environment Output: Amount of the phenomena of interest at location Amount of information at a location is uncertainty about output value Select points in environment with highest uncertainty (entropy) Brad Woosley – Research and Creative Activity Fair 2017 6

Communications modeling GP Input: 2 D - Point in environment Output: Signal Strength at location Learn layout of network signal strengths as move through environment Select points from GP that have highest signal strength Brad Woosley – Research and Creative Activity Fair 2017 7

Selection of best point Brad Woosley – Research and Creative Activity Fair 2017 8

Reducing communications Not all collected information immediately useful to all other robots Usefulness decreases with distance between robots Ad-Hoc network, spatially close robots are linked via communications Form communication tree Track area bounded by robots under that node Convex hull Check for going to same point by traversing down tree Notify of new point selected by going up the tree Brad Woosley – Research and Creative Activity Fair 2017 9

Ongoing work Implementation on ROS using turtle bot Testing Obstacle layouts Distributions of information Communication range Data to collect Time taken Accuracy of model generated Distance traveled Brad Woosley – Research and Creative Activity Fair 2017 10

Conclusions Proposed approach for multi-robot exploration of unknown environment while considering information collection and communication constraints Applications to extraterrestrial exploration and disaster recovery Brad Woosley – Research and Creative Activity Fair 2017 11

Questions? Contact: Brad Woosley bwoosley@unomaha. edu Brad Woosley – Research and Creative Activity Fair 2017 12