The Mobile Agents Architecture supporting Science
Extra-Vehicular Activities on Planetary Surfaces
Maarten Sierhuis, Ph.D.
Project Leader for the Mobile Agents Project
RIACS/NASA Ames Research Center

Introduction:

We are back at the MDRS. A year and a half ago we left this Mars habitat in the Utah desert. As Rotation 16 we tested our Mobile Agents Architecture for two weeks inside and outside the habitat. It was exhilarating to see how more than 20 researchers came together to test their software and hardware for supporting planetary science on Mars. This year we are the last crew of the season, Rotation 29. We are again here to test out our software and hardware.

Our systems have improved. We fixed problems, added new features based on ideas and requirements gathered last year, and we have expanded the project to include collaboration with a remote science team distributed around the world. This year we will test how teams back on Earth can participate in planning EVAs and analyze the science data that the astronauts and their robotic assistant gather during the EVA.

We have increased our capability such that we now also support parts of the EVA planning and science analysis work process on Mars and on Earth. Our software agents take care of most of the data flow in the process. Plans developed by the crew and remote science team is automatically loaded into the Mobile Agents Architecture and distributed to all the software agents that need this plan to perform their support roles during an EVA.

Here we describe the software architecture and tools that are currently part of what we call the Mobile Agents Architecture. As this research progresses based on out field test experience, the Mobile Agent Architecture changes. New capabilities are added to the existing software agents, new software agents are added, new software systems are integrated, new speech dialog capabilities are needed, et cetera. Following is a description of what we will test this year.

It is useful to provide a definition of what we mean with a software agent, because this term is used in many different ways in computer science. Software agents are a rather new phenomenon in computer science. Some call it a new software design paradigm. If we look at the benefit of our software agents in the design and implementation of the Mobile Architecture, we would agree with calling software agents not simply a way of programming, but a new paradigm for software development.

Software agent: A software component with which humans or other agents can interact as if it is an independent behavioral entity.

The Mobile Agents Architecture (MAA) consists of four software foundations, the Java programming language, the Brahms multiagent modeling and programming language, the KAoS agent-communication framework, and Corba. Together these software tools allow us to implement the MAA distributed over a wide variety of wirelessly networked computers, from the astronaut spacesuit computers, to the robot computers and the computers in the MDRS habitat, to the computers for the Remote Science Team (RST) back on Earth.

Figure 1. Mobile Agents elements - Click for Detail

Figure 1 shows a graphic of the current hardware, network and software elements in the MAA. Starting at the right side of the graphic we see the two EVA astronauts with their computational capability. The computer inside the astronaut's backpacks runs the Brahms software agents that integrate and communicate with all the other software pieces in the architecture and coordinate the dataflow between the systems. The dialogue system software allows the astronauts to speak to their personal software agents using human speech. The GPS system gathers GPS data from the 2 cm accurate differential backpack GPS units. The biovest system gathers astronaut health information from health-monitoring sensors. Digital cameras can be connected to the backpack computers so that software agents can download digital image files taken by the astronaut and associate them with other gathered data (such as the astronaut's GPS data), as well as store and forward these images to the habitat and Earth.

Next, one of the ERA robot computers runs the Brahms software agents supporting the ERA robot in executing plans and allowing the robot to communicate its information to the Brahms software agents on the astronaut's spacesuit and in the habitat. The ERA software agents allow us to easily integrate the ERA robot's capabilities with the rest of the software in the architecture. The ERA robot can be used in two different modes, an autonomous mode and an interactive team mode. In the autonomous mode an ERA software agent executes a previously uploaded EVA plan. This allows the robot to go on EVAs by itself and take panorama and other images that are immediately communicated back to the habitat. In the interactive team mode the ERA supports the astronauts on their RVA and gets its commands either via a speech command from the astronauts or habitat communicator (Habcom), or from a graphical user interface (GUI) operated by the Habcom person back in the habitat. The ERA can take digital panorama images, find its way to named GPS locations, carry the astronaut's tools and science samples, print a curation label for the rock and soil samples gathered by the astronauts and autonomously follow an astronaut. The ERA robot can also function as a network relay node for the astronauts, allowing the astronauts to wonder to interesting areas providing network connectivity back to the habitat.

An all-terrain vehicle (ATV) serves as a wireless network "hub" between the astronauts, the ERA robot and the habitat. Using this approach we can reconfigure the communication path for the dataflow between astronauts, robot and habitat. The ATV computer serves as the MAA agent directory service, allowing the agents to find each other on this distributed network. The KAoS software (http://www.ihmc.us) that is part of the Brahms agent software environment in effect creates a distributed agent framework allowing the MAA software agents to be distributed and communicate over a 5 km long wireless WAN.

Inside the MDRS habitat a set of software agents serve as the central coordination, communication and storage facility integrated with a number of other software systems previously developed to store science data, and allow distributed teams to collaborate asynchronously (this is important because the communication time-delay between Earth and Mars is too long to have synchronous collaboration going on). This integrated habitat software environment allows the crew to collaboratively, but asynchronously, create EVA plans with the remote science team back on Earth, as well as analyze science data in the habitat after an EVA. Software agents running inside the habitat automatically monitor and store science and EVA plan data that is communicated by the software agents of the astronauts and ERA robot. After storing this data, a software agent automatically e-mails EVA data to the RST back on Earth via a satellite connection to the internet. At the same time the data stored inside the habitat's ScienceOrganizer system (http://sciencedesk.arc.nasa.gov) is automatically copied to a ScienceOrganizer database back on Earth. This allows the RST to access the data in near real-time (currently we do not incorporate a communication time-delay into the system). EVA plans are created and communicated by the crew to the RST using the Compendium software tool (http://www.compendiuminstitute.org) that is specifically developed for capturing meetings. This Compendium tool is used to allow both the crew and the RST to collaborate as distributed science teams on two different planets. A Compendium software agent in the habitat can read the EVA plan from the Compendium database and distribute the plan amongst all the software agents in the MAA. This agent also stores all science data that comes into the habitat in the Compendium database, so that the crew and the RST can discuss the data within the Compendium tool.

Figure 2 gives a graphical view of the software agents part of the MAA. The green icons circles in the three circles shows the software agents running inside the Brahms virtual machine on the astronaut's backpack, the ERA robot and on the Habcom computer inside the hab.

Figure 2. Brahms software agents in the Mobile Agents Architecture - Click for Detail