3D Simulation of Oil Concentration from a Leak

Information Based Search in Turbulent Flows

Variations in material concentration resulting from a biochemical or radiological contaminant leakage, such as an oil spill in the ocean or a radioactive dispersal in the atmosphere, is dominated by turbulent mixing.  The result is a highly anisotropic and unsteady sensory landscape where sensor measurements become the sporadic and intermittent which renders gradient based search strategies highly ineffective.   This work develops information based search strategies for autonomous robots to search and localize the source of a biochemical contaminant dispersed in turbulent media.  The approach has been validated using state-of-the-art 3D computational fluid models of the 2010 Deep Water Horizon oil spill developed by Dr. Alex Fabregat Tomás at CUNY.


Distributed Sensing and Tracking of Geophysical Flows

Geophysical fluid dynamics (GFD) is the study of natural large-scale fluid flows, such as oceans, the atmosphere, and rivers. GFD flows are naturally stochastic and aperiodic, yet exhibit coherent structure. Coherent structures are important because they enable the estimation of the underlying geophysical fluid dynamics.   While these transport controlling structures in GFD flows are inherently complicated and unsteady, their understanding is necessary for the design of robust underwater vehicle control and the prediction of various physical, chemical, and biological processes that in GFD flows.  Nevertheless, the data sets that describe GFD flows are often finite-time and of low resolution and most transport controlling features in fluids are unstable and non-stationary, this renders the problem of “mapping” these features using teams of autonomous vehicles highly challenging.

The goals of this project are to overcome the theoretical and technical challenges and develop a general mathematical and control framework for distributed autonomous sensing and tracking of geophysical fluid dynamics in 2D space over time (2D+1) and in 3D space over time (3D+1).  The key idea exploits the capability of the team to cover large regions in physical space to increase the spatio-temporal sampling resolution of the flow field. The data will then be processed in a distributed fashion by the team to obtain a global description of the flow dynamics that can be maintained and updated in real time.  Objectives of this work include:

  1. Identifying and evaluating key kinematic features that control transport in oceanic surface flows of greatest relevance to autonomous vehicle navigation and control;
  2. Synthesizing distributed sensing and tracking strategies for a class of transport controlling features in 2D, 2D+1, 3D, and 3D+1 flows;
  3. Developing a general mathematical and control framework for teams of autonomous vehicles that leverages key transport controlling features in oceanic flows for improved navigation and monitoring of dynamic and uncertain environments;
  4. Developing a new control and coordination framework that leverages the disparate sensing, mobility, and computational capabilities within a heterogeneous teams of mobile robots for ocean monitoring;
  5. Experimentally validate the proposed strategies via a proof-of-concept robotic system with simulated and realistic flow data.


Two Robots Straddling a Manifold in the Multi-Robot (MR) Tank

Two robots driving on opposite sides of the stable manifold of saddle point in an actual time-invariant flow created in the MR tank. The robots are obtaining measurements of the flow field using their onboard flow sensors which are capable of measuring both the speed and direction of the current relative to the robot’s movements. Global position for the robots are provided using an external OptiTrack motion capture system.

Tracking Lagrangian Coherent Structures

Three-robot manifold/LCS tracking strategy validated on a periodic wind-driven double-gyre model, experimental data, and actual ocean data.

Full simulation run three robots tracking the LCS off the coast of Santa Barbara, CA.


  1. D. Kularatne and M. A. Hsieh. “Tracking Attracting Manifolds in Flows,” Autonomous Robots, 41(8), Mar 2017, pp. 1575–1588. Bibtex | PDF
  2. M. Michini, M. Ani Hsieh, E. Forgoston, and I. B. Schwartz. “Robotic Tracking of Coherent Structures in Flows,” IEEE Trans. on Robotics, Vol. 30, No. 3, 593-603, Jun 2014. Bibtex | PDF
  3. D. Kularatne and M. A. Hsieh. “Tracking Attracting Lagrangian Coherent Structures in Flows,” in the 2015 Robotics: Science and Systems (RSS 2015), July 2015, Rome, Italy. BibTeX | PDF
  4. M. Michini, M. A. Hsieh, E. Forgoston, and I. B. Schwartz, “Experimental Validation of Robotic Manifold Tracking in Gyre-Like Flows,” in the IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS’14), Sep 2014, Chicago, IL USA. BibTeX | PDF
  5. M. Michini, H. Rastgoftar, M. A. Hsieh, and S. Jayasuriya, “Distributed Formation Control for Collaborative Tracking of Manifolds in Flows,” in the 2014 American Control Conference (ACC 2014), Jun 2014, Portland, Oregon USA. BibTeX | PDF

Task Partitioning for Distributed Assembly

Distributed autonomous assembly of general two (2D) and three dimensional (3D) structures is a complex task requiring robots to have the ability to: 1) sense and manipulate assembly components; 2) interact with the desired structure at all stages of the assembly process; 3) satisfy a variety of precedence constraints to ensure assembly correctness; and 4) ensure the stability and structural integrity of the desired structure throughout the assembly process. While the distributed assembly problem represents a class of tightly-coupled tasks that is of much interest in multi-robot systems, it is also highly relevant to the development of next generation intelligent, flexible, and adaptive manufacturing and automation. In this work, we address the assembly of a three dimensional structures by a team of robots. Specifically, we address the challenges of partitioning a complex assembly task into N loosely coupled tasks each executed by a single robot. Furthermore, we have developed online sensing capabilities that enable the team to determine the state of the structure during assembly to allow for the identification and correction of assembly errors.


Autonomous Assembly

Two mobile manipulators assembling a 3-D structure. A planning algorithm partitions the assembly task into N subcomponents that can be executed by individual robots with minimal communication with other robots. See IROS 2011 paper for details.

Robotic assembly of Lincoln-log styled parts.

Municipal Park in Belo Horizonte, Brazil by James Milligan

Multi-Robot Systems for Large Scale Cooperative Tasks

This project provided undergraduate and graduate students a unique opportunity to work with an interdisciplinary and international team of researchers on the design and control of multi-agent robotic systems. The technical focus of the collaboration was centered around the design of robust multi-robot coordination strategies for execution of large scale cooperative tasks. Advances in embedded processor and sensor technology in the last thirty years have accelerate the demand for teams of robots in various application domains. Multi-agent robotic systems are particularly well-suited to execute tasks that cover wide geographic ranges, require significant parallelization, and/or depend capabilities that are varied in both quantity and difficulty. Example applications include littoral exploration and surveillance, rainforest health monitoring, autonomous transportation systems, warehouse automation, and hazardous waste clean-up.

Grant: NSF OISE 113011.

Project Alumni

Daniel Mox, B.S./M.S. 2015, Drexel University, Ph.D. student at the University of Pennsylvania
Dennis Larkin, M.S. 2015, Drexel University
Emily LeBlanc, B.S. 2014, Temple University, Ph.D. student at Drexel University
James Milligan, B.S./M.S. 2013, Drexel University, SRI