Development of a Flexible Command and Control Software Architecture for Marine Robotic Applications

and the scientific community have been early adopters of this technology, guiding the development of new capabilities to satisfy the needs of applications such as ocean mine countermeasures (Alt, 2003; Schnoor, 2003; Eickstedt & Schmidt, 2003; Rish et al., 2001), seafloor survey for oil and gas exploration (Chance et al., 2000; Vestgard et al., 1999), and studying deep-sea hydrothermal vents (Yoerger et al., 2007a, 2007b; Sohn, et al., 2008; Kunz, et al., 2009). More recently, autonomous vehicles have been used for disaster response such as responding to oil spills similar to the recent Deepwater Horizon incident (Camilli et al., 2010; Bhattacharya et al., 2011) and natural disasters such as hurricanes (Murphy et al., 2008, 2009). The focus of this article is on the development and implementation of a flexible software architecture based around the lightweight communications and marshalling (LCM) message passing protocol. We report on the development of the command and control software developed for two marine robot applications: supervisory control of a maritime domain awareness (MDA)mission and themodificaMay/J tion of an Ocean-Server Iver2 autonomous underwater vehicle (AUV) for real-time visual feature-based navigation. This software is based on the open-source LCM project initiated to support the Defense Advanced Research Projects Agency (DARPA) UrbanGrandChallenge. The architecture we propose has distinct advantages over some of the available software tools for robotics, but it is not a unique solution. We describe a generic framework for this software along with benchmarking tests to compare the performance of this system with other robotics operating tools. une 2011 Volume 45 Number 3 25 Background and Closely Related Work Real-Time Command and Control A fundamental aspect of robotic software is the use of modularity to cope with the complexity of real-time operation. For example, each hardware sensor typically has a software driver that manages the sensor communication and passes the data from each sensor to estimation, control, and planning software modules. Generically, the process of communicating between modules is called interprocess communication (IPC). The choice of IPC method is the key to the performance of the real-time robotics software. The LCM system is a relatively new alternative for implementing IPC for real-time robotics in the marine environment (Huang et al., 2009, 2010). LCM was developed specifically for low-latency, high-throughput communication to support the MIT entrant in the DARPA Urban Grand Challenge (Leonard et al., 2008). The authors of LCM have compared their design with similar systems currently in use by the robotics community including the Player project, the CARMEN robotics package, the Joint Architecture for Unmanned Systems, the Robotics Operating System (ROS), Microsoft’s Robotics Development Studio, and the Mission Oriented Operating Suite (MOOS) (Huang et al., 2009). LCM provides a set of software tools for message passing between modules (processes or threads). These tools provide the A “message” is any digital data that can be encapsulated into a C-like data structure. For example, messages could be simple text, images, files, or binary data. 26 Marine Technology Society Journa following functions for the development of a real-time system: ■ message type specification for l anguage i ndependen t da t a structures; ■ marshalling (encoding and decoding) of LCM messages via software libraries for C/C++, Java, Python, MATLAB, and C#; ■ s ca l ab le communica t ion v ia user datagram protocol (UDP) multicast. The design of LCM emphasizes the high-throughput, low-latency needs of distributed robotics applications and accomplishes this by providing the infrastructure that holds together the modules of a software solution rather than prescribing the structure of any of the processes or threads. The result is an extremely general software architecture that is amenable to the needs of a supervisory control scheme described below. The LCMmessages are transmitted across the network via multicast UDP packets. In this method, a process publishes the data on a particular channel, defined by a unique name. Any other process on the network can subscribe to the same channel and receive the UDP message. The marshalling capabilities encode and decode the structured data transmitted over the network. The fingerprinting functions of LCM guarantee that the messages are type-safe, i.e., that the communicating processes/threads agree exactly about how to interpret the messages. This design allows for transparent distribution of the system across a standard Ethernet or wireless network. The MOOS software system is being applied to a growing number of marine robotic applications. In contrast to the LCM libraries, which provide a minimal message passing functionality, MOOS is a more coml plete set of communicating applications including both the libraries and executables for a real-time platform (Newman, 2003). Communication is done in a publish/subscribe model, referred to as a star topology where each client subscribes to a central database where all messages are stored. Message passing is done using human-readable text strings containing one or more name/value pairs via transmission control protocol (TCP). Typically processes and threads that interact with the database are derived from a common application. The central database can then control the message passing and execution of software developed for the particular platform. MOOS has been used for a variety of ocean robotics applications including AUVs and unmanned surface craft (Curcio et al., 2008). The MOOS software has more recently been extended to support cooperative control of multiple underwater platforms (Benjamin et al., 2010; Tye et al., 2009). One of the goals of this research is to qualitatively and quantitatively compare the MOOS system with an LCM implementation of real-time robotic command and control (Table 1). Unmanned Vehicles for Port and Harbor Security Unmanned marine vehicles (both surface vehicles and underwater vehicles) continue to evolve as important components of defense, industry, and scientific ocean operations. As these platforms have matured, they have found application in an increasingly wide variety of operational scenarios. The research presented here is a part of a burgeoning impetus to apply robotic technology to MDA. One aspect of MDA that has seen considerable attention is the need for unmanned systems to assist in disaster response, including search and rescue. A number of commercial platforms (e.g., the iRobot Packbot and FosterMiller Talon) are available for urban search and rescue (USAR). This technology has aided a number of important USAR incidents, the World Trade Center disaster being a particularly dramatic example (Casper & Murphy, 2003). Unmanned aerial vehicles (UAVs) have also been tasked with disaster response such as surveying structural failures like the Berkman Plaza II collapse in 2007 (Pratt et al., 2008). Similarly, UAVs have been applied to wilderness search and rescue (Goodrich, et al., 2008). These examples demonstrate the capability of unmanned systems for security applications using terrestrial and aerial applications, but this potential is yet to be realized by the maritime security community. In the marine environment, researchers have leveraged the robotic technologies developed for defense for maritime event response. The Center for Robot-Assisted Search and Rescue has deployed both Unmanned Surface Vehicles (USVs) and UAVs for postdisaster port and littoral inspection after hurricanes Ike andWilma (Steimle et al., 2009;Murphy et al., 2008, 2009). The authors’ use of the acronym “UPSV” is not meant to add a new acronym to the long list of acronyms in the literature, but this acronym is used for brevity in this article. Supervisory Control Supervisory control explicitly includes the human operator in the feedback loop of an automated system. Thomas Sheridan coalesced much of the early thought on this paradigm, “supervisory control means that one or more human operators are intermittently programming and continually receiving information from a computer that itself closes an autonomous control loop through artificial effectors to the controlled process or task environment” (Sheridan, 1992). The framework we have developed is inspired by Sheridan’s emphasis on careful attention to the capabilities of the human and robotic components of such a system. Figure 1 illustrates the supervisory architecture where the human supervisor can interact with the system at a variety of levels. By building this flexibility into the software, we can provide users with an interface that allows them to simultaneously automate tasks and take direct control of the platform depending on the situation. Case Study I: University of Hawaii Field Robotics Laboratory We present the design of an unmanned system for port and harbor security as part of the Department of Homeland Security’s (DHS) initiative to enhance MDA. At the Field Robotics Laboratory (FRL) at the University May/J of Hawaii, we have designed, built, and tested the unmanned port security vessel (UPSV) to investigate the unique challenges of the homeland security mission. The UPSV concept is designed around the need for port resilience after a transportation security incident (TSI). A TSI can be the result of a natural disaster (e.g., hurricane, earthquake, or tsunami), environmental damage (e.g., chemical spill), or terrorist attack. As an example of the potential impact, U.S. West Coast ports support 8 million U.S. jobs as 14 million TEUs (20-foot equivalent units) are moved through these ports (Annual Report, 2009). Because of the economic and security risks associated with the impact of such an event, there TABLE 1 Comparison of LCM and MOOS communication implementations.