Multi-robot goal conflict resolution under communication constraints using spatial approximation and strategic caching

https://doi.org/10.1016/j.robot.2020.103713Get rights and content

Abstract

We consider the problem of distributed goal conflict resolution in multi-robot systems while remaining resilient to intermittent communication losses between robots. Our proposed approach uses a spatial approximation technique called α-shape to represent the regions that have been explored by robots followed by a O(logn) algorithm that incrementally combines and shares the α-shape information between robots along the robots’ communication tree and rapidly checks for conflicts of a robot’s selected location. We provide theoretical guarantees of the time complexity of our proposed algorithm along with experimental results with simulated and physical robots in different environments. The results show that our approach can rapidly determine conflicts between goal locations selected by multiple robots as well as reduce message loss and re-transmissions between robots. These result in more efficient inter-robot communications as well as less extraneous distance traveled by robots, as compared to a flooding-based communications approach.

Introduction

Multi-robot exploration for information collection with distributed, networked robots is an important task in many robotic applications including unmanned search and rescue, robotic reconnaissance, perimeter surveillance and robotic detection of physical phenomena such as radiation and underwater concentration of lifeforms [1], [2], [3]. An essential aspect in distributed multi-robot information collection is to ensure that a task performed by a robot is not repeated by other robots, resulting in deterioration of the robots’ performance. For example, multiple robots repeatedly sampling information from the same locations could result in unnecessary energy, time and communications expenditure. The problem is further complicated as robots might have to intermittently leave the communication range of other robots or a base station while exploring a larger environment, and, during these intermittent periods, they have to make decisions and perform actions independently without sharing information or coordinating with other robots. To address these issues, we propose a method for de-conflicting goal locations that multiple robots may want to visit. In particular we look at the case where a set of robots, in a distributed manner, select goal locations to visit from a large set of possible locations, and need to ensure that the selected location, or a near by one, has not already been selected by another robot. Our proposed technique uses a geometric approximation called an α-shape [4] to group together regions of the environment that a robot can communicate with other robots using multi-hop communications over a communications network. This technique is integrated with an intelligent search algorithm over the robots’ communication tree to find conflicts, and store them even if the robot that selects the goal disconnects from the communication tree before reaching the goal. To the best of our knowledge, our work is one of the first attempts to integrate geometry-based prediction of potential conflict regions to improve multi-robot information collection under communication constraints, while gracefully handling intermittent connectivity loss between robots. We have validated our algorithm’s worst case and average case communications properties analytically. We have also reported experimental results on simulated robots within multiple environments and physical Clearpath Jackal Robots. Our results show that the α-shapes based region approximation technique can significantly reduce overlap between regions of multiple robots as compared to coarser approximation techniques like the convex hull. Combined with the intelligent search algorithm over the robots’ communications tree the technique can successfully improve communications throughput by up to 34% as compared to a flooding based approach; sends up to 62% fewer messages than the flooding approach, and, reduces the percent extraneous distance traveled by robots by 0.29% to 34% in simulation and between 15% and 19% in hardware tests compared to the flooding approach.

The rest of this paper is structured as follows: we discuss related works in Section 2. In Section 3, we describe our region approximation technique using α-shapes and corresponding conflict resolution algorithm. Theoretical results on the time complexity of our proposed algorithm are discussed in Section 4. Section 5 describes our experimental results while evaluating the effectiveness of our α-shapes-based conflict resolution and communications algorithm on both simulated and physical robots, and, finally, we conclude.

Section snippets

Related work

Autonomous multi-robot based systems have recently been proposed for collecting information from environments about ambient phenomena such as temperature, nuclear radiation or microscopic lifeforms from underwater, using a technique called Multi-robot Information-driven Path Planning (MIPP) [5]. MIPP uses Gaussian Processes (GP) [6] to build a model of the spatial distribution of the phenomena of interest using mutual information entropy between different locations in the environment and direct

Model

We consider a set of robots R operating inside a bounded environment QR2, divided into two sets QobsQ that represents all regions blocked by obstacles, and Qfree=QQobs represents all regions that are in free space. Qobs and Qfree are initially not known to the robots and they discover Qobs as they navigate within the environment. Each robot rR selects a goal location qrQfree to which they wish to navigate. Our main objective is to develop an algorithm which can check if a location qr

Theoretical results

Theorem 1 Worst Case Time Bound

The worst case time bound on finding a conflict in the tree, if it exists is: O(|Vt|) where |Vt| is the number of robots in the current communications tree.

Proof

In the worst case, assume a branching factor b=1, meaning that each node has only one child. Also, assume that the search starts at a leaf node and the conflict is stored in the root node. To reach the root, all |Vt| nodes must first be checked. Once the conflict is detected, then the information must be relayed back to the leaf node,

Experimental results

We have evaluated our proposed α-shapes based goal location conflict resolution approach within a multi-robot information sampling scenario inside an initially unknown environment, using both simulated and physical robots. Our experiments were designed around testing the following hypotheses to evaluate the performance of our proposed approach:

  • H1: Using α-shapes instead of coarser spatial representations such as convex-hulls improves the precision of approximated regions by reducing the overlap

Conclusions and future work

We proposed and evaluated a method for resolving conflicts between a set of robots that independently select locations to visit in a distributed information sampling application. Experimental evaluations of our approach showed that it significantly reduces the amount of communications required and the amount of extraneous distance the robots traveled as compared to a flooding approach and less extraneous distance than an auction approach, and, that robots could gracefully leave and join the

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Bradley Woosley is a Ph.D. candidate in the Computer Science Department at the University of Nebraska at Omaha (UNO). He received his M.S. degree from UNO in computer science in 2015, and his B.Sc. in Computer Engineering from the University of Nebraska- Lincoln in 2013.

References (37)

  • RasmussenCarl Edward

    Gaussian Processes for Machine Learning

    (2006)
  • CaoNannan et al.

    Multi-robot informative path planning for active sensing of environmental phenomena: A tale of two algorithms

  • Sampling-based robotic information gathering algorithms

    The International Journal of Robotics Research

    (2014)
  • CuiRongxin et al.

    Mutual information-based multi-AUV path planning for scalar field sampling using multidimensional RRT*

    IEEE Trans. Syst. Man Cybern. Syst.

    (2016)
  • BinneyJonathan et al.

    Optimizing waypoints for monitoring spatiotemporal phenomena

    Int. J. Robot. Res.

    (2013)
  • LiceaD. Bonilla et al.

    Mobility diversity-assisted wireless communication for mobile robots

    IEEE Trans. Robot.

    (2016)
  • Y. Solana, M. Furci, J. Cortés, A. Franchi, Multi-robot path planning with maintenance of generalized connectivity, in:...
  • S. Kemna, J.G. Rogers, C. Nieto-Granda, S. Young, G.S. Sukhatme, Multi-robot coordination through dynamic Voronoi...
  • Bradley Woosley is a Ph.D. candidate in the Computer Science Department at the University of Nebraska at Omaha (UNO). He received his M.S. degree from UNO in computer science in 2015, and his B.Sc. in Computer Engineering from the University of Nebraska- Lincoln in 2013.

    Prithviraj Dasgupta is a research scientist at the U. S. Naval Research Laboratory doing research in the areas of adversarial machine learning, multi-agent systems, game-playing AI and game theory. From 2001 through 2019 he was the Union Pacific Endowed Professor in the Computer Science department at the University of Nebraska at Omaha where he founded and directed the C-MANTIC Robotics lab. He has led multiple, large, federally-funded projects in the area of multi-robot/ multiagent systems and published more than 150 research papers in leading conferences and journals in the areas of multi-robot and multi-agent systems. He received his Ph.D. in 2001 from the University of California, Santa Barbara.

    John G. Rogers III is a research scientist specializing in autonomousmobile robotics at the Army Research Laboratorys Information Sciences Division (ISD) of the Computational and Information Sciences Directorate (CISD). At ARL, John is leading an effort to enable mobile robot information gathering activities for Warfighter mission support through a decision process which considers the valueof the information being collected. John completed his Ph.D. degree at the Georgia Institute of Technology in 2012 with his advisor, Prof. Henrik Christensen from the Robotics and Intelligent Machines center.

    Jeffery Twigg received his Bachelors and Masters degrees from the Virginia Polytechnic Institute and State University in Engineering Mechanics. He joined the Army Research Laboratory in 2010. Currently, he is a Computer Engineer and Roboticist in the Intelligent Robotics Branch of the Computational and Information Sciences Directorate at ARL. He researches control strategies for maintaining and optimizing connectivity between robotic agents which use WIFI and low-VHF bands for communication. He is also pursuing a PhD in mechanical engineering at University of Maryland.

    This research was partially supported by the Army Research Laboratory, USA and was accomplished under Cooperative Agreement Numbers W911NF-19-2-0235. Prithviraj Dasgupta performed his part of this research while he was a professor in the Computer Science Department at the University of Nebraska at Omaha. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the Army Research Laboratory or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes not withstanding any copyright notation herein.

    View full text