1. Introduction

The past decade has seen plans to send a manned mission to Mars by about 2033 (NASA 2004). The NASA 2003 Strategic Plan stated that "NASA will continue to expand its human presence in space — not as an end in itself, but as a means to further the goals of exploration, research, and discovery" (NASA 2003). In January 2004, NASA established a long-term program to extend human presence across the solar system, with a primary goal to establish a human presence on the Moon no later than 2020, as a precursor to human exploration of Mars (NASA 2004). Given human missions to Mars it is important to consider the enormous stress on the human beings who will embark on this incredible journey of exploration. Every effort must be made to ensure the astronauts' safe return (or there will never be public support to send them out), their well-being, and their effectiveness in exploring this new world. Although humans may never discover ways to practically or safely go farther than Mars, we have had the capability to develop the means to reach Mars for decades.

A vision to visit Mars must have both a motivational "why?" component; and a technological "how?" component. The motivational component is tied to its geology, atmosphere, and water. We contend that the technological component should include an appropriate combination of robotic autonomy and human telesupervision.

2. Why Go to Mars?

Mars is an extremely interesting place. The Red Planet has water (much of which is now frozen, but had very likely flowed in the past) and a thin atmosphere (some of which freezes every winter, and is composed primarily of carbon dioxide with less than three percent nitrogen, compared to Earth's, which is over three-quarters nitrogen). It has a significant amount of sunlight, and a well-defined surface. Compared with the other planetary bodies in our solar system, Mars is the only one that can be described as "somewhat hospitable". And it is nearby, relatively speaking.

We've been to our Moon, and for decades there has not been enough interest to have motivated decision makers to fund a return. There are many very worthwhile things to do on the Moon, but Mars is the much more compelling destination. We will go to Mars. The question is not "Can we?", for we can; rather, the question is "When will we have the will?" Developing this will involves casting a coherent and compelling vision and, on the strength of this vision, marshaling the required economic and human resources.

3. Telesupervision

Planetary surfaces are among the most hostile working environments for humans. The protective clothing required (a spacesuit) is massive and cumbersome, and limits Extra-Vehicular Activity (EVA) times because all life-sustaining resources (air, water, energy, etc.) must also be carried. The costs to lift and supply habitation for each human are tremendous. While many tasks may be managed entirely by autonomous software agents, humans must still be present to plan, monitor, and act in the more complicated tasks. By off-loading those tasks or parts of tasks for which robotic technology is most suited, humans will be far more effective.

Robots can extend operational capabilities and robustness in the face of harsh environmental conditions, whereas humans possess varied expertise and the ability to correctly interpret and react to novel situations difficult to achieve in robots. Both human safety and robot utility are increased when combining these advantages in a human-telesupervised system. Employing multiple robots to perform dangerous and repetitive tasks with maximal autonomy preserves human attention resources and allows humans to focus on high-level guidance and assistance when situations are beyond the capabilities of the autonomy. Such a system thus multiplies both human and robotic capabilities.

A telesupervision system requires an architecture integrating several key technologies: humanrobot interaction with variable autonomy; robot team task allocation and cooperation; and highfidelity telepresence. Such a system forms an essential part of a viable technological approach for the exploration of Mars. A structured approach to appropriate technology application is supported by our experiences in developing multi-robot telesupervision systems for planetary mineral prospecting (Podnar et al. 2007); for ocean algal bloom investigations (Podnar et al. 2008); and for surface water quality assessment (Podnar et al. 2010).

By taking best advantage of robotic systems, we can multiply astronauts' effectiveness by more efficiently applying their time, and by reducing EVAs and their associated safety risks. There is also a concomitant reduction of hundreds of tons of mass that would otherwise need to be transported to Mars by reducing astronauts' support requirements to complete the exploration tasks. Astronaut effectiveness and safety are further increased by improving the astronaut's work environment: working from a "shirtsleeve environment" base (Fig. 1), rather than from a cramped vehicle and performing many EVAs.

In the present article, the advantages and limitations of current Mars exploration technologies are first described; our proposed telesupervisory approach is detailed next; and finally the technology research, development, and testing needed to support this vision of exploration are described.

4. Current Technologies

The first Mars rover, Sojourner, a magnificent culmination of efforts in the Pathfinder mission over a decade ago, demonstrated one method of remotely exploring another planet (Mishkin 1998, Taylor 2007). The two rovers that landed on the surface of Mars in early 2004, Spirit and Opportunity, have expanded on these operations with astounding successes, including combined traverses of 30km in six years (NASA 2010). Both then and now, the exploration carried out by these vehicles has been neither autonomous nor teleoperated, but primarily "batch autonomy" remote human control: a team of "rover drivers" prepare the moves for the next Martian day, planning the path from the images downlinked from the previous day. Limited autonomous driving has been allowed in 0.5-2 meter steps (NASA 2010). Unfortunately, the technologies required to allow such robot vehicles to explore with significant autonomy, depending on little or no human intervention, remain relatively immature and unproven in space mission environments.

Significant research in the area of human supervisory control of multi-robot systems has been performed in laboratory and field environments on Earth for a range of applications, including construction (Fong 2006; Simmons 2007), search and rescue (Nevatia 2008;Wang 2009), hazardous materials detection (Bao 2010), navigation (Crandall 2005, Touvain 2006), and exploration (Podnar 2006, 2008). Fong et al. (2006) present a Human-Robot Interaction Operating System designed to enable humans and robots to engage in task-oriented dialogue and problem-solving. Simmons et al. (Simmons 2007) emphasize high-granularity sliding autonomy and heterogeneous robot teams. Crandall and Goodrich (Crandall 2005) use the concepts of neglect time (the time a robot is effective between required human interventions) and interaction time (the time a human needs to restore a robot to effectiveness) to predict how many robots a human can supervise. Wang et al. (2009) point out the deficiency of these measures for strongly cooperative tasks and propose differentiated human roles as a way of handling increased control difficulty with growing robot team size. Trouvain et al. (2006) use multi-modal visual and auditory feedback to improve telesupervisory control. Key results of this research are the recognition of the value of shared human-robot autonomy to system performance and the analytical and experimental characterization of its effects. Shared autonomy has various names in the literature, including sliding, adjustable, variable, and augmented autonomy, but the basic idea is that the degree of autonomy the robots exercise can vary along a spectrum from none (when a human teleoperates) to full (when humans simply monitor).

The scientific success of planetary missions depends on having humans in the loop. However, this carries with it several disadvantages: the amount of time for collection of scientific data is small compared to the time spent waiting for path planning and batch programming (by humans on Earth); limited communication 'windows' (twice per Martian sol); and significant communication delays. Because we must have Earth-bound mission controllers to initiate actions and receive data returned, and because the maturity and mission acceptance of autonomous technologies are limited, this long-distance exploration is extremely slow. Sending human beings to Mars will certainly provide a much greater science return because the time "loop" will be much tighter. However, human safety is paramount and more difficult to assure in this case.

5. Proposed Telesupervision Approach

Human control of robotic assets can be implemented on a variety of levels: from direct human remote control with no autonomy, to full autonomy with no human control. Figure 2 illustrates the era in which direct remote control, or telerobotics, had significant development (1950s- 1980s), and when fully autonomous systems were first being developed (1980s-1990s). Telesupervision, incorporates both telerobotics and robotic autonomy, and supports a full range of levels from high-level planning, through levels of semi-autonomous operation, down to direct human telecontrol.

The remainder of this section describes the key components of our approach to telesupervisory human control of multi-robot systems for Mars exploration: 1) a telesupervision architecture for shared autonomy, planning, communications, and visualization; 2) human supervision from Mars orbit rather than on its surface; and 3) high-fidelity telepresence for human fault detection and science analysis.

5.1 Telesupervision Architecture Our multi-level autonomy telesupervision architecture provides an integrated approach to multirobot coordination and multi-level robot-human autonomy. (Fig. 3) It allows multiple robotic assets (both mobile and fixed) to function in a cooperative fashion, and the operating mode of different robots to vary from full autonomy to teleoperated control.

High-level Task Planning and Monitoring allows a human telesupervisor to assign to a fleet of robotic assets high-level goals, such as specifying an area of ground or section of cliff to investigate, which are then automatically subdivided into operational commands sent to each robot by the Robot Team Coordinator. As the robots execute these plans their operation is monitored both by the Robot Team Coordinator module and by the human telesupervisor. Adaptive replanning of the robot assignments is based on sensor inputs (dynamic sensing) and coordination between multiple assets, thereby increasing data-gathering effectiveness while reducing the human effort required for tasking, control, and monitoring of the vehicles. Multi-level autonomy includes: low-level autonomy on each independently-operating robot; autonomous monitoring of the fleet; adaptive replanning; and when necessary, intervention by the human telesupervisor either with manual replanning, or by taking direct control of a robot via teleoperation. This full range of autonomy is supported by a hazard and assistance detection capability that has two basic roles: detection of the need for human intervention and management of the human attention resource. Robots use probabilistic methods (Elfes 1996) to assess environmental obstacles or objects of potential scientific interest and determine whether navigational or interpretive assistance is required. The resultant requests for human intervention are evaluated, prioritized, queued, and scheduled (Mau & Dolan, 2007) using the concepts of estimated "interaction time" (the time needed for human intervention to handle a robot request) and "neglect time" (the time after which a robot is likely to require assistance again) (Goodrich & Olsen, 2003). This methodology permits a multi-robot team to pursue mission accomplishment as rapidly and safely as possible without unduly burdening the operator. Algorithms for science analysis of the acquired data can perform an initial assessment of the presence of specific science signatures of immediate interest both onboard each robot, and at the telesupervisor's workstation. These data and assessments can be shared with distant experts (with the concomitant communications delays) for further analysis.

Figures 4, 5 and 6 depict an experiment in which a human telesupervisor at Carnegie Mellon University in Pittsburgh, Pennsylvania (Fig. 4) specified a high-level goal of mineral detection to a Sample Return Rover (SRR) operating at the Jet Propulsion Laboratory (JPL) in Pasadena, California using Internet communications and camera-based visual feedback. The upper sequence of images in Fig. 5 show the view of the telesupervision workstation screen with its controls and remote visual feedback, while the lower sequence shows the SRR at JPL. The SRR was capable of obstacle avoidance and safeguarding that prevented the telesupervisor from damaging it. The autonomy software (whose monitoring screen is shown in Fig. 5) was additionally able to indicate to the telesupervisor the likely presence of a rock, which the telesupervisor could then examine and manipulate using direct teleoperation (Fig. 4).

5.2 Human Telesupervision from Mars Orbit When we eventually land human beings on Mars, many of the planetary operations will be best carried out by a few humans who spend most of their time telesupervising multiple robots from a "shirtsleeve environment" base as has been described. However, with Mars surface gravity 38% that of Earth, the energy to safely land capable systems, supplies, and the humans that will use them is enormous. Then we must add to this the costs of landing the vehicles and fuel to return the humans to Martian orbit. An estimated cost-per-kilogram to deliver functional equipment to the surface of Mars is a staggering $309,000 (Mitchell 2008-2009). As a result, no matter how symbolically satisfying it would be, it is likely beyond the economic scope of any near-term effort to land human beings on the surface of Mars and to return them safely. However, it is enormously less expensive to transport human beings into orbit around Mars and support them there, and to land only robotic assets that need not be lifted back into Martian orbit.

By avoiding landing humans on Mars (on the first trip), we also dramatically decrease mission complexity, and thus increase the probability of their safe return. By reducing the mass needed to be transported to Mars possibly by orders of magnitude, achieving a human mission to Mars with a high-value science return can be within our economic means far sooner. Therefore, one can justify an initial program for more direct human exploration of Mars, gaining almost all the science return of having humans walk on the surface without the need to land them. But this does require having human beings present in Mars orbit.

5.3 High-Fidelity Telepresence One of the greatest frustrations of scientists is the inability to "be there". The roundtrip communication delays (Fig. 7) from Earth to Mars (6-45 minutes) severely limit today's science return. We propose landing robotic exploration vehicles on Mars that have the capability of providing high-fidelity telepresence to human beings who will be much closer (1/8-second roundtrip delay) in Mars areostationary orbit. This is practically as good as "being there" without compounding the risks by trying to fund, build, transport, land, assemble, test, and successfully use a complete human spaceflight center on Mars.

The more realistic the experience of presence is, the more effective the remote scientist, or engineer, or technician can be. Realistic telepresence requires high-fidelity remote reproduction of visual, tactile, kinesthetic, and aural sensations. Geometrically-correct, high-resolution stereoscopic video (Grinberg et al. 1994, Grinberg et al. 1995) supports faithful visual reproduction. Force feedback manipulators support important aspects of kinesthetic proprioception, i.e., the sense of the relative positions of neighboring parts of the body. Tactile sensor arrays coupled to tactile reproducers "take the gloves off" of remote fine manipulation tasks. And stereophonic audio, even in a thin atmosphere, can provide a richness of information about the mechanical interface between hand-and-tool, tool-and-workpiece, and body-andworksite. The temporal experience is also a significant factor in faithful sensory reproduction (Matteazzi 2009). The longer the round-trip communications delay, the slower each human action-reaction becomes, until certain tasks become impractical. This temporal limitation on effective telepresence circumscribes the distances over which a full telepresence system is useful: optimally within a communications sphere defined by a roundtrip of 1/6 light second (well beyond areostationary orbit). Therefore, locating a human base in areostationary orbit and landing only robotic assets allows these robots to be effectively telesupervised, both by monitoring autonomous operations, and by taking direct control of a robot when necessary with a realistic sense of being there through high-fidelity telepresence.

6. Technology Prospectus

As an appropriate precursor to the proposed approach to a first human Martian exploration mission, it is prudent to deploy and test systems in easier-to-reach but similarly challenging space environments. We propose Earth orbit and the Moon as appropriate testbeds for proving technologies that will then be deployed for human exploration of Mars. The potential costs and risks of conducting Mars exploration are so great (especially in human life) that an integrated and thorough research, development, and testing approach is crucial to success and sustained operations.

7. Developing an Earth Orbit Telepresence

In addition to preparing for successful human exploration of Mars, there are also near-term benefits to developing and proving the capabilities for telepresence between the surface of the Earth and satellites on orbit. If we had the ability to support human beings long term in low Earth orbit (not only from within the confines of the International Space Station) and geostationary orbit we could:

These tasks typically require human presence, and have therefore been cost-prohibitive. As these places are also nearby (within the optimal temporal telepresence sphere when supported by a low-latency surface-to-orbit communications system) and of great commercial interest, they are ideal places to exploit telepresence technologies to support humans on orbit remotely.

Communication satellites can be less expensive, as the entire satellite need not be engineered to withstand the forces of lift. Less expensive components can be safely packed for lifting and assembled on orbit. Servicing of equipment already on orbit multiplies its value by extending its useful life, and preventing it from becoming more dangerous space junk.

Experts across many more fields can be remotely present, and all without needing to qualify as astronauts. And while only one person can perform manipulative tasks with each system, more than one expert can share the same view, thus providing additional benefits of having mentors. Further, while astronauts need to be provided for when they are resting as well as working, telepresence hardware can be employed around the clock by work shifts of Earth-based experts.

8. Semi-Autonomous and Telepresence Systems for Lunar Exploitation

With many astronauts working long hours through telepresence systems, optimizing their effectiveness and well-being is paramount to successful and sustained missions. The Moon is about two-and-a-half seconds minimum communications roundtrip from the Earth; therefore, full telepresence is not suitable for Lunar operations. Autonomous tasks with remote oversight are possible with limited direct intervention. This semi-autonomous approach can be very effective for such repetitive tasks as prospecting, mining, and systematized construction. However, if we have a human presence on the Moon, full telepresence will be a great multiplier for the effectiveness of each person. For instance, there is no need for anyone in a spacesuit to drive a bulldozer or erect an antenna, when instead they can reside in a "shirtsleeve" environment of a permanent base, and remotely control equipment of all kinds. Servicing is far better accomplished via high-fidelity telepresence, which again removes the need for working from inside a bulky spacesuit, with limited consumables.

Employing this human-and-robot telepresence technology supports longer EVA work sessions with better fine motor control. Periodic travel to and from a worksite which is remote from a base is also eliminated. Safety of the humans living on the Moon is also significantly enhanced. Thus, this Lunar deployment of a telesupervision system is a very reasonable analogue for a human Martian exploration mission.

8. Summary

Telesupervision of multi-robot systems has great potential to increase science return, and human safety and productivity when we send humans to explore Mars. Appropriate sharing of risks and workload among humans and robots will allow human attention to stay focused on critical tasks. Our experience in developing a telesupervision architecture for Lunar and Martian exploration using a homogeneous fleet of mineral prospecting robots (Podnar et al. 2006; Halberstam et al. 2006; Elfes et al. 2006; Dolan et al. 2005) demonstrated some of these benefits. The experiences gained from this initial NASA-supported work, as well as a project on telesupervision of multiple surface craft for Harmful Algal Bloom detection (Podnar et al. 2008), have allowed development and testing of some of the enabling technologies, including high-level mission planning, hazard and assistance detection, and high-fidelity telepresence including geometrically-correct stereoscopic remote vision systems.

Augmenting human abilities with robotic systems is crucial to achieving the long-term goals of human space exploration. Employing the described telesupervision architecture provides formal integration of multi-robot coordination and multi-level robot-human autonomy. Initially deploying humans on orbit rather than on the Martian surface will increase safety while allowing round-trip communications times compatible with effective telesupervision. Testing the technology in low Earth orbit and on the Moon prior to Mars will contain risk and costs.