Metareasoning for Robots and Autonomous Systems

This page contains the following sections:

• Background
• Metareasoning Approaches

Principal Investigator

• Jeffrey W. Herrmann

Book: Herrmann, Jeffrey W., Metareasoning for Robots: Adapting in Dynamic and Uncertain Environments, Springer Nature, 2023.

Figure caption: Possible runs of agents performing decentralized TA for each type of scenario considered in our problem formulation. Each target has a unique ID, not shown in these figures to reduce clutter. In the ship protection scenario, lighter shades of green reflect that a longer time has passed since cells were last searched. Image: Carrillo et al., 2021.

Collaborators and students: Shapour Azarm, Michael W. Otte, Mumu Xu. Tekevwe Akoroda, Akshay Bapat, Estefany Carrillo, Mohamed Khalid M Jaffar Sharan Nayak, Ruchir Patel, Eliot Rudnick-Cohen, Suyash Yeotikar.

Related Publications

• Nayak, Sharan, Suyash Yeotikar, Estefany Carrillo, Eliot Rudnick-Cohen, Mohamed Khalid M Jaffar, Ruchir Patel, Shapour Azarm, Jeffrey W. Herrmann, Huan Xu, Michael W. Otte, Experimental Comparison of Decentralized Task Allocation Algorithms under Imperfect Communication, IEEE Robotics and Automation Letters, DOI: 10.1109/LRA.2019.2963646, published online: January 2020; video.
• Patel, Ruchir, Eliot Rudnick-Cohen, Shapour Azarm, Michael W. Otte, Huan Xu, Jeffrey W. Herrmann, Decentralized Task Allocation in Multi-Agent Systems using a Decentralized Genetic Algorithm, 2020 IEEE International Conference on Robotics and Automation (ICRA), Paris, May 31-June 4, 2020: video.
• Samuel T. Langlois, Oghenetekevwe Akoroda, Estefany Carrillo, Jeffrey W. Herrmann, Shapour Azarm, Huan Xu, and Michael Otte, Metareasoning Structures, Problems, and Modes for Multiagent Systems: A Survey, IEEE Access, October 5, 2020. DOI: 10.1109/ACCESS.2020.3028751
• Estefany Carrillo, Mohamed Khalid M. Jaffar, Sharan Nayak, Ruchir Patel, Suyash Yeotikar, Shapour Azarm, Jeffrey W. Herrmann, Michael Otte, and Huan Xu, Communication-Aware Multi-Agent Metareasoning for Decentralized Task Allocation , IEEE Access, Volume 9, pp. 98712-98730, 2021, doi: 10.1109/ACCESS.2021.3096229

Metareasoning for Thermal Management

This project conducted a study that evaluated how well metareasoning can manage a Raspberry Pi 4B's central processing unit (CPU) temperature while it is performing image processing (object detection and classification) on the Common Objects in Context (COCO) dataset. We developed and tested two metareasoning approaches: the first maintains constant image throughput, and the second maintains constant expected detection accuracy. The first approach switched between the InceptionV2 and MobileNetV2 image classification networks with a Single Shot Multibox Detector (SSD) attached. The second approach was tested on each network for a range of parameter values. The study also considered cases that used the system's built-in throttling method to control the temperature. Both metareasoning approaches were able to stabilize the device temperature without relying on throttling.

Michael Dawson conducted this study, which was funded by the Army Research Laboratory as part of the ArtIAMAS cooperative agreement.

Figure caption: Metareasoning can modify how quickly an image processing algorithm operates to avoid excessive temperature. SSD image: Liu et al., ECCV 2016.

Related Publications

• Dawson, Michael K., Metareasoning Approaches to Thermal Management During Image Processing, M.S. Thesis, University of Maryland, College Park, 2022.
• Dawson, Michael, and Jeffrey W. Herrmann, Metareasoning Approaches for Thermal Management During Image Processing, DETC2022-88459, Proceedings of the ASME 2022 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference (IDETC/CIE2022), St. Louis, Missouri, August 14-17, 2022.

Using Metareasoning on a Mobile Ground Robot to Recover from Path Planning Failures

Autonomous mobile ground robots use global and local path planners to determine the routes that they should follow to achieve mission goals while avoiding obstacles. Although many path planners have been developed, no single one is best for all situations. This paper describes metareasoning approaches that enable a robot to select a new path planning algorithm when the current planning algorithm cannot find a feasible solution. We implemented the approaches within a ROS-based autonomy stack and conducted simulation experiments to evaluate their performance in multiple scenarios. The results show that these metareasoning approaches reduce the frequency of failures and reduce the time required to complete the mission.

Sidney Molnar, Matt Mueller, Robert Macpherson, and Lawrence Rhoads conducted this study, which was funded by the Army Research Laboratory as part of the ArtIAMAS cooperative agreement.

Related Publications

• Molnar, Sidney, Matt Mueller, Robert Macpherson, Lawrence Rhoads, and J.W. Herrmann, Using Metareasoning on a Mobile Ground Robot to Recover from Path Planning Failures, Institute for Systems Research, University of Maryland, College Park, February 16, 2023.

Figure caption: Simulated robot stuck behind a cluster of trees, where its planner is unable to create a feasible path. Image: Sidney Molnar.

Data-driven Metareasoning for Collaborative Autonomous Systems

When coordinating their actions to accomplish a mission, the agents in a multi-agent system may use a collaboration algorithm to determine which agent performs which task. This paper describes a novel data-driven metareasoning approach that generates a metareasoning policy that the agents can use whenever they must collaborate to assign tasks. This metareasoning approach collects data about the performance of the algorithms at many decision points and uses this data to train a set of surrogate models that can estimate the expected performance of different algorithms. This yields a metareasoning policy that, based on the current state of the system, estimated the algorithms’ expected performance and chose the best one. For a ship protection scenario, computational results show that one version of the metareasoning policy performed as well as the best component algorithm but required less computational effort. The proposed data-driven metareasoning approach could be a promising tool for developing policies to control multi-agent autonomous systems.

This study was conducted during my sabbatical at the Naval Air Warfare Center-Aircraft Division (NAWC-AD) in 2018. I would like to thank Stephen Kracinovich, Johann Soto, and their colleagues at NAWC-AD for the financial support and the resources used in this research.

1. Herrmann, Jeffrey W., Data-driven Metareasoning for Collaborative Autonomous Systems, Technical Report, Institute for Systems Research, University of Maryland, College Park, January, 2020.