Advancing the Scientific Frontier with Increasingly Autonomous Systems
Rashied Amini, Abigail Azari, Shyam Bhaskaran, Patricia Beauchamp, Julie Castillo-Rogez, Rebecca Castano, Seung Chung, John Day, Richard Doyle, Martin Feather, Lorraine Fesq, Jeremy Frank, P. Michael Furlong, Michel Ingham, Brian Kennedy, Ksenia Kolcio, Issa Nesnas, Robert Rasmussen, Glenn Reeves, Cristina Sorice, Bethany Theiling, Jay Wyatt
AAdvancing the Scientific Frontier with IncreasinglyAutonomous Systems
Rashied Amini , a , Abigail Azari , Shyam Bhaskaran , Patricia Beauchamp , Julie Castillo-Rogez ,Rebecca Castano , Seung Chung , John Day , Richard Doyle , Martin Feather , Lorraine Fesq ,Jeremy Frank , P. Michael Furlong , Michel Ingham , Brian Kennedy , Ksenia Kolcio , IssaNesnas , Robert Rasmussen , Glenn Reeves , Cristina Sorice , Bethany Theiling , Jay Wyatt Endorsements
Tibor Balint , Luther Beegle , Steve Chien , Chad Edwards , Carolyn Ernst , Terrance Fong ,Abigail Fraeman , Anthony Freeman , Carly Howett , Robert Lillis , Gentry Lee , MichaelMischna , Marc Rayman , Mark Robinson September 15, 2020 NASA Jet Propulsion Laboratory Space Sciences Laboratory, University of California, Berkeley NASA Ames Research Center Okean Solutions NASA Goddard Space Flight Center Johns Hopkins University Applied Physics Laboratory Southwest Research Institute Arizona State University a [email protected], (626) 720-9942c (cid:13) a r X i v : . [ a s t r o - ph . I M ] S e p Executive Summary
A close partnership between people and partiallyautonomous machines has enabled decades ofspace exploration. But to further expand ourhorizons, our systems must become more capa-ble. Increasing the nature and degree of auton-omy - allowing our systems to make and act ontheir own decisions as directed by mission teams -enables new science capabilities and enhances sci-ence return. The 2011 Planetary Science DecadalSurvey (PSDS) and on-going pre-Decadal mis-sion studies have identified increased autonomyas a core technology required for future missions.However, even as scientific discovery has neces-sitated the development of autonomous systemsand past flight demonstrations have been success-ful, institutional barriers have limited its matura-tion and infusion on existing planetary missions.Consequently, the authors and endorsers of thispaper recommend that new programmatic path-ways be developed to infuse autonomy, infrastruc-ture for support autonomous systems be investedin, new practices be adopted, and the cost-savingvalue of autonomy for operations be studied.
From the beginning of interplanetary exploration,reliance on on-board decision-making has beencritical for mission success. The use of time-driven command sequences and critical sequenceretries on the Viking and Mariner 6 & 7 orbitersresulted in reduced risk and increased the abil-ity of spacecraft to return science data [1, 2].Over the following decades, the evolutionary in-clusion of autonomous functions, primarily in thedomains of spacecraft fault protection, and guid-ance and control, further reduced mission risk andincreased a spacecraft’s ability to perform anddownlink science measurements. For instance,the Jovian radiation environment caused multiplesafe mode events during the Galileo mission; but“smart safing” enabled it to maintain thermally-safe attitude to protect its instruments despiteimmediate loss of operator control. However, themission and system complexity needed to answernew questions in planetary science has outpacedefforts to mature autonomous capabilities. As de-tailed in this paper, many of the ambitious mis-sion concepts described in the Planetary MissionConcept Studies (PMCS) for the Planetary Sci- ence and Astrobiology Decadal Survey (PSADS)and the 2018 Workshop on Autonomy for FutureNASA Science Missions will not be achievable us-ing the current paradigm of spacecraft control [3,4].To answer ever more challenging science ques-tions, we will need spacecraft that can exploreunknown and dynamic environments with less in-put from human operators. This will demand anintegrated approach to autonomy, because auton-omy is a system-level technology that requiresan interdisciplinary approach to technology de-velopment. The need for a revolutionary ad-vance in spacecraft autonomy to meet the needsof NASA’s missions was identified as early as the1980 Carl Sagan, et al. report to NASA on Ma-chine Intelligence and Robotics [5]. Yet despitethe need for and past uses of autonomous capa-bilities, these capabilities are rarely used or evenconsidered in most planetary missions due to in-stitutional barriers.In the Sagan Report, one of the primary barri-ers to progress in autonomous systems was iden-tified as culture: “Technology decisions are, tomuch too great a degree, dictated by specific mis-sion goals, powerfully impeding NASA utilizationof modern computer science and technology. Un-like its pioneering work in other areas of scienceand technology, NASA’s use of computer scienceand machine intelligence has been conservativeand unimaginative.”
Similar cultural issues wereidentified at ESA, based on the experience of 2019OP-SAT mission: “...resistance to experimenta-tion and innovation, especially when the projectedbenefits are not yet flight proven. Time aftertime projects settle for reuse rather than inno-vation.” [6] To wit, the highest level recommen-dation of this paper is one made by the SaganReport: “The advances and developments in ma-chine intelligence and robotics needed to make fu-ture space missions economical and feasible willnot happen without a major long-term commit-ment and centralized, coordinated support.”
The 2015 NASA Technology Roadmap defines au-tonomy as “the ability of a system to achievegoals while operating independently of externalcontrol.” [7] External control is exemplified bythe traditional commanding control loop, sum-marized in Figure 1. Autonomy reduces the char-1 igure 1:
The traditional commanding control loophas a characteristic timescale that may be greater thanthe timescales needed to exercise control. In thesecases, autonomous functions are required. acteristic control timescale by moving analysisand planning on board the spacecraft [8]. Mov-ing functions traditionally performed by missionteams to on board the spacecraft requires changesto both flight systems and ground systems.When the timescale associated with traditionalcommanding becomes larger than timescales re-quired for responding to spacecraft critical eventsor interacting in an unpredictable environment,using a priori planning for control increases mis-sion risk. Deploying autonomy, in situ percep-tion, reasoning, and acting under both nominaland off-nominal situations, mitigates this risk andallows the spacecraft to make decisions based oncurrent circumstances.Commanding timescales are partly driven bycommunications for commands and status. Forinstance, in the case of planetary rotorcraft likeIngenuity or Dragonfly, communication is con-strained by limited data rates (e.g. due toweak signal), restricted link availability (e.g. abody’s orbit and day/night cycle), and largedata products required for ground-based plan-ning (e.g. contextual maps for path planning)[9]. The timescale of systems responding tocritical events or performing critical operationsis driven by the dynamical nature of system-environment and system-system interactions andhow predictable and observable these interactionsare. Dynamic rotorcraft mobility in planetary at-mospheres is incompatible with communicationconstraints and has to occur in situ .It is this relationship between communications-and dynamical-related timescales that has tra-ditionally driven requirements for autonomy.Where the communications timescale exceeds thedynamical timescale, safe control using strict
Figure 2:
Autonomy is necessary when a spacecrafthas to react to changes in the environment or withinitself at a shorter timescale than afforded by communi-cation constraints. Past missions relied on their abil-ity to predict but future missions that operate in in-creasingly unknown environments would fall outsidethe controllability threshold. a priori planning cannot be assured and au-tonomous functions are required for safe control.Figure 2 illustrates the control regimes and as-sociated risks of using a priori planning. It’sfor this reason that autonomous functions werefirst used for mission aspects with short dynam-ical timescales, e.g. guidance, control, and faultprotection [2]. With incremental confidence fromthese early missions, more ambitious autonomousfunctions were flown to enable more ambitiousscience measurements. For example, in 2004Deep Impact’s Impactor spacecraft required au-tonomous navigation functions (AutoNav) to in-tercept Comet Tempel 1 [10]. This was driven bytwo primary goals: first, colliding with the cometusing the Impactor spacecraft, and second, ob-serving the impact on the Flyby vehicle. Becausethe exact size, shape, and orbit of the comet couldnot be determined in advance from the ground,closing the navigational loop on-board was nec-essary to successfully impact the 7 km size cometat a speed of 10 km/s. The use of AutoNav wasunavoidable to meet the science objective andAutoNav’s decision-making could be sufficientlymodeled and supervised that it could be trusted.While using autonomy may be enabling forsome missions, it can enhance mission produc-tivity and science return for nearly all missions,increasing a mission’s science-to-dollar ratio. On-board science data analysis has been used on var-ious Mars rovers [11, 12] and on Earth Observing-1 [13] to improve science return and dynamicallytarget and image novel signatures. For instance,2uriosity incorporated Autonomous Explorationfor Gathering Increased Science (AEGIS) to tar-get the ChemCam laser-induced breakdown spec-troscopy (LIBS) instrument, targeted preferredoutcrop terrain over 93% of the time as opposedto blind targeting, which targeted outcrops 24%of the time [14]. Two additional white paperssubmitted to the PSADS, Azari et al and Theil-ing et al, further describe the advantages of usingscience autonomy [15, 16].However, a lesson learned from Mars rovers,EO-1, Deep Impact, and Stardust is that flight-proven and high value-to-risk-ratio autonomousfunctions are not regularly used on Flagship orcompeted missions. Moreover, these deploymentshave not led to the development of system-levelautonomy, which is able to integrate numerousautonomous and traditional functions.
GivenNASA’s track record of developing and de-ploying autonomous systems and the inter-action of autonomous functions necessaryfor strategic planetary science missions, itis clear a shift in institutional culture isnecessary.3.1 Future Planetary Missions and TheirAutonomy Requirements
All of the missions under evaluation by thePSADS are either enabled by, or would be en-hanced by, the use of increasingly autonomoussystems. By evaluating aspects of mission ar-chitecture that constrain communications or aredriven by the dynamics of system-system andsystem-environment interactions we can identifyhow these missions are enabled and enhanced.These evaluations are summarized in Table 1, andadditional description for two mission types areincluded below.
Landed Missions (landers, hoppers, rovers,aerial systems)
Landed systems are impacted bysimilar issues that drive the use of autonomy.Across these classes of missions, the dynamic as-pects of operations during entry, descent, andlanding (EDL), roving, flying, and hopping re-quire situational awareness and in situ reason-ing and acting. Limited line of sight to Earthand communication constraints restrict contactopportunities and data rates. Limited lifetimeand data volume constraints, such as a Europaor a Venus lander, will need situational aware-ness, assessment of the interaction with the un- known surface for sampling, and “smart” target-ing to increase science return and reduce risk un-der off-nominal conditions [19]. On-board anal-ysis of “remote”, low-cost data, like Raman orLIBS, can significantly reduce risk of performinghigh-cost sampling, like drilling [20, 21, 22].Just as critical to these time-limited missionsis a capability for restorative fault managementthat can restore functionality after a safing event,or take action to avoid the need for a saf-ing event. As more decision-making capabil-ity is moved on-board, the scope of autonomousfault detection, isolation and recovery (FDIR)functions will increase, relying on on-board re-planning/scheduling and execution of real-timecontingency actions. This sort of functionality isbroadly applicable to science missions operatingin all contexts but critical to missions with signif-icant and challenging operations outside availablecommunication windows. (E.g. Mercury Landernighttime operations with six weeks of no groundcontact. Also, Intrepid operations that have tocover an 1,800 km distance with hundreds of in-strument placements in four years, where manualinterventions cannot occur more than once every6–16 km of traverse, or missions seeking to accessextreme surface or subsurface terrains. [23, 24,25, 26])Generally, by adopting an approach wheresituational awareness (perception, mapping, es-timation, see [27]), hazard assessment, plan-ning/execution, payload data analysis, scienceplanning, and FDIR functions are moved on-board, landed missions will be more produc-tive. As an example, results from the Self-ReliantRover study, wherein a terrestrial rover was oper-ated by campaign intent rather sequenced activi-ties, showed an 80% reduction in sols required tocomplete a campaign and 267% increase in loca-tions surveyed per week [28].
Deep Space Missions
Although deep-space pro-vides a more predictable environment than thatnear, on, or into planetary bodies (rendezvous,proximity operations, surface mobility, below-surface access), autonomy enables operationswith reduced communication burden and allowsscaling to multi-craft missions for deep-space (or-biters and flybys). Using autonomous naviga-tion functions provides tighter turnaround loopsin high dynamic environment situations wherelong light-times precludes ground processing to3 able 1:
Summary of how different aspects related to mission architecture can be enabled or enhanced throughthe use of integrated autonomous functions.
Aspect of Mission Architecture Drivers Enabling/EnhancingAutonomous FunctionsEntry, Descent, and Landing , e.g.- Mercury Lander, Enceladus Orbilander
Surface & Aerial Mobility , e.g.- Intrepid Rover- Lunar/Vesta Geochronology Hopper- Dragonfly, Mars Helicopter - Short dynamic timescales- Limited a priori atmosphere/surfacecharacterization- One-way light times (OWLTs) - Terrain-relative navigation (TRN)and dynamic control- Hazard assessment and avoidance- Payload data analysis- Planning/execution- Restorative fault management
Short-Lived Landers , e.g.- Venus Flagship lander- Europa Lander- VIPER Rover - Science competitiveness- Limited lifetime- Limited bandwidth and contact opportunity - Payload data analysis- Planning/execution- Restorative fault management
Missions with Opportunistic Science
Potentially all missions, but particularly:- Fast flyby missions, e.g. [17]- Monitoring missions, e.g. MOSAIC- In-situ missions, e.g. Ceres Habitability - Limited communications and time toperform critical targeting- Science competitiveness and impact- Cost/risk reduction - Autonomous navigation (enablingfor fast flybys)- Planning/execution- Science planning- Payload data analysis
Interplanetary Cruise , e.g.- All missions beyond Earth-Moon system- Missions using electric propulsion, e.g. Persephone - Mitigating impact of safing on trajectory(EP missions)- Cost/risk reduction - Autonomous navigation- Planning/execution- Science planning- Payload data analysis
Missions with Coordinated Observations , e.g.- Multi-SC mapping missions, like MOSAIC- Bistatic radar experiments, e.g. CONCERT- Planetary defense & impactor/observer missions, e.g.DART, Small Bodies DRM [18]- Landed system coordination, e.g. M2020/MarsHelicopter, lava tube exploration - Limited time to coordinate with ground-in-the-loop- Science competitiveness- Cost/risk reduction - Multi-agent coordination- Planning/execution- Science planning- Autonomous navigation
Mapping Missions , e.g.- Europa Clipper- MORIE - Mitigating the impact from anomaly- Science competitiveness- Cost/risk reduction - Planning/execution- Science planning
Operations in High-Radiation Environments , e.g.- Europa Clipper- Io Volatiles Explorer - Communications delay to restore scienceoperations may not be acceptable- Reduce risk/cost - Restorative fault management
Approach/Rendezvous with Unexplored Bodies , e.g.- NEO/NEA missions [18]- Comet Sample Return- Transneptunian bodies/KBO etc. - Uncertainty in relative spacecraft/body position- Unknown irregular body shape and gravity- Unknown geotechnical properties for landing- Limited a priori surface characterization - Autonomous navigation- Autonomous mapping- TRN, hazard assessment, landing- Autonomous surface navigation- Restorative fault management achieve required accuracy. Autonomy can en-able operations in less predictable scenarios, likeatmospheric aerocapture at icy giants, whereturnaround time on ground-based navigation mayinduce additional risk, and for planetary constel-lation where coordinated, multi-spacecraft oper-ations are required [29, 30].As the sensitivity of ground-based surveys im-prove over the next decade, the detection fre-quency of interstellar visitors, e.g., ‘Oumuamuaand 2I/Borisov, and long-period comets, likeC/2017 K2, is expected to increase [31]. At thesame time, numerous proposals have called forflybys of distant objects such as Trojans, Jovianand Saturnian moons, trans-Neptunian objects,and Kuiper Belt objects. Both sets of missionsface similar challenges: flybys of these bodies in-volve high relative velocities, limiting the effec-tiveness of ground-based navigation and scienceplanning to the point where the mission may notbe feasible. Integrating autonomous navigation,payload data processing, and planning/executionfunctions may be enabling for these missions [17]. In architectures with multiple spacecraft, e.g. atwo spacecraft flyby of a NEO for bistatic radarinvestigation, multi-agent coordination, naviga-tion, and planning/execution can be used to per-form coordinated measurements to achieve chal-lenging measurement objectives and maximizepayload utilization based on available resources.The Mercury, Venus, Ceres, MORIE, MO-SAIC, and Persephone PMCS reports have base-lined electric propulsion [24, 32, 33, 34, 35]. Elec-tric propulsion trajectories are non-Keplerian anduse continuous thrusting to benefit from theirhigh I SP . However, safe mode events can resultin missed thrust, risking mission success in termsof schedule and excessive propellant use to cor-rect trajectory. For Dawn, a four-day period ofmissed thrust resulted in a 26-day delay to thefirst planned Ceres orbit; a projected seven-daymissed thrust could have resulted in a ∼ There are also institutional motivations for us-ing autonomous functions on-board spacecraft.A shift in culture that is more accepting of au-tonomous systems would see its benefits in relax-ing constraints on deep space communication andscheduling, mission competitiveness, and achiev-ing more with the planetary exploration budget.
Deep Space Network & Communications In-frastructure
All missions utilizing the Deep SpaceNetwork (DSN) could be expanded and see im-provements in science return using adaptive op-erations. The introduction of deep space Small-Sat missions is affecting the roadmap for NASA’sDeep Space Network (DSN) and is leading toa fundamental change in the way future deepspace missions will interact with ground systems[38]. With the projected increase in the num-ber of DSN users, e.g. SmallSats funded throughSIMPLEx, the DSN will need to more efficientlyschedule tracking passes. Increasing the degree ofspacecraft autonomy will allow improvements inthe efficiency of DSN use.
Mission Competitiveness
The interrelated fac-tors of cost, perceived risk, and science returnimpact selection of competed missions. Inte- grating different autonomous functions can affordPIs more flexibility in meeting science require-ments, relaxing system requirements on missionsystems, and potentially reducing cost and sci-ence risk. For instance, using autonomy like au-tonomous targeting can enable performing noveland opportunistic measurements, raising a mis-sion’s threshold science return. Use of auton-omy like autonomous retry with on-board plan-ning/execution, can reduce mission risk by allow-ing a spacecraft to dynamically adopt to on-boardanomalies and reattempt measurements.A notable example of a mission that could be-come more competitive should systems autonomybe used is a New Frontier Venus lander - a con-cept that has seen multiple failed proposals overtwo decades, including the step-two Venus In situComposition Investigations (VICI) and the VenusSurface and Atmosphere Geochemical Explorer(SAGE) proposals [39, 40]. By integrating theautonomous functions in Table 1, a Venus lan-der could more effectively utilize its time beforeend-of-mission. Contextual images and remoteRaman/LIBS measurements can direct sampling,resulting in lower risk on science return. Based oncurrent development of the NASA ARC VolatilesInvestigation Polar Exploration Rover (VIPER),this could enable higher performance and betterhandling mission-ending faults.
Cost
Autonomy could reduce costs to NASAby moving some of operations team functions onboard. In the last 10 years, SMD has spent$2.4B FY20 on mission operations [41]. If thesemissions had been operated 20% more efficiently,$480M could have been available to NASA. Aspredicted in the 1980 Sagan Report, moving to“Autonomous Mission Control” by the year 2000would result in mission operations costs 1% ofthose in 1975. While “Autonomous Mission Con-trol” has yet to be realized, the experience frompast missions and current research supports thatprediction. On EO-1, significant mission time wassaved by having the spacecraft discard imageryobscured by cloud cover.
Despite decades of incremental progress in achiev-ing remarkable successes with the autonomousfunctions used on EO-1, different Mars rovers,and missions like Deep Impact and Stardust, thebroad use of autonomous functions to achieve5ission objectives is still not standard practice.While the reasons are interconnected, they canbe generalized as three barriers to infusion.
Barrier 1: Unlike most other NASAtechnology investments, autonomy issystem-level technology.
Most technologies,e.g. detectors and propulsion systems, fulfillspecific needs. Their operational conditionscan be readily modeled for laboratory testingand interfaces and behavior well-defined formission development and operations. For thesetechnologies, the NASA definitions of TechnologyReadiness Level (TRL) are a relevant and usefultool for appraising flight-readiness. Moreover,incremental investments that raise TRL are ap-propriate. Autonomy, especially where multiplefunctions are integrated, has implications for thesystem’s architecture, design, development andoperations processes, and personnel throughout.It requires coordinated efforts throughout devel-opment phases and across domains and cannotbe comprehensively adopted through incrementalinvestments. This means standard paths tomature technologies do not apply to autonomoussystems. Experience has shown that technologydemonstration missions, like the Remote AgentExperiment on Deep Space-1, do not transitionto routine science mission use [42].
Barrier 2: Institutional environment lim-its autonomy to incremental maturationand thus restricts NASA’s ability to deployautonomous systems.
Even though autonomyhas the potential to reduce mission risk, usingit is still a perceived risk. NASA has success-fully executed missions without significant auton-omy for decades and is capable of developing andevaluating missions that are enhanced or enabledby autonomy. As exemplified by AutoNav andAEGIS and described in the Sagan Report andESA OPS-SAT paper, a cultural barrier limitsopportunities to fly enhancing technologies. Riskaversion results in incremental maturation of spe-cific autonomous functions, which struggle againto find their place on science missions despite pastsuccess and potential advantages. Even whenrisks are taken on competed missions and theywould stand to benefit in science-to-dollar ratio,e.g. Dragonfly, opportunistic science is not con-sidered an intrinsic component of baseline mis-sions. As also noted in the Sagan Report and stilltrue today, NASA struggles to pioneer software advances. The nature of the competitive missionprocess drives engineers at NASA and in industrytoward heritage solutions in mission proposals toavoid perceived risk. Integrated over time, thissteers missions away from innovative advancesthat could be enabling for more demanding mis-sions. There is little interdisciplinary coopera-tion to offer guidance on performing software- andautonomy-related trade studies, which could havesignificant implications for mission architecture.This lack of exposure and strained mapping ofTRL to multi-mission autonomous functions fur-ther reinforces a culture wary of adopting moreadvanced autonomy.
Barrier 3: Lack of Inter-Directorate Co-ordination.
The Human Exploration Direc-torate (HEOMD), the Space Technology Mis-sion Directorate (STMD), and divisions of Sci-ence Mission Directorate (SMD) have struggledto coordinate and lack direct incentive to developNASA-wide plans to implement autonomous sys-tems. This results in fragmented and incremen-tal progress that will not lead the agency to sus-tainable advances in multi-mission autonomy. Al-ternatively, coordinated investments could allowdirectorates to leverage developments by others,promoting agency-wide sharing of standards andpractices, increasing the likelihood autonomousfunctions are reflown, and minimizing duplicatedinvestments. Meanwhile, private industry bene-fits from investments in autonomy and can pushforward with increasingly autonomous solutions,e.g. the Falcon fly-back boosters used by SpaceX.
Given how autonomy would enable and enhancestrategic missions of interest to the planetary sci-ence community, we offer the following high-levelrecommendations for enabling the routine deploy-ment, and continued evolution, of autonomy forfuture planetary science missions. These high-level recommendations include several potentialimplementations described as examples.1.
Create programmatic pathways thatprepare integrated autonomy systemsfor future missions and build institu-tional trust, e.g.: • Commit to advancing autonomy by settinga series of capability deadlines to includeincreasing amounts of autonomy on all6lanetary science missions. This would en-sure that NASA is ready with the neededprocesses and capabilities when the timecomes to fly the more ambitious missionsthat are enabled by autonomy • Incentivize adoption of autonomous sys-tems for PSD Announcements of Opportu-nity (AOs), e.g. through a cost incentive. • Coordinate STMD investments with com-peted missions, e.g. SIMPLEx, so missionspush boundaries of science exploration andtechnology demonstration • Expand programs like the New FrontiersHomesteader Program and ROSES to in-clude autonomous functions, system inte-gration, operator tools, and verificationmethods for autonomous systems. (SeeRec • Instrument AOs for Flagships should offeran opportunity for collaborative propos-als with complementary instruments usingautonomy for payload data processing andscience planning • Use all extended missions for demonstrat-ing autonomous functions2.
Invest in infrastructure for developingand supporting autonomous systems inspace, e.g.: • With inter-directorate coordination, investin an in-space autonomy testbed, poten-tially utilizing SmallSats, so NASA cen-ters and industry can test and flight vali-date flight and ground software and trainon new processes (See Rec • Make DSN and Advanced Multi-MissionOperations System (AMMOS) invest-ments that support the anticipated growthof customer missions and use of au-tonomous systems • Expand Homesteader and ROSES op-portunities to cover interdisciplinary au-tonomous research and development (SeeRec • Set specific objectives and a time frame fortransitioning AMMOS away from a priori planning (e.g., change from time-based se-quencing to goal-based commanding). Seta date where all new missions would beexpected to use this new paradigm. 3.
Invest in practices that promote themulti-mission use of autonomous sys-tems. Practices includes design, devel-opment, test, verification, and opera-tions processes and standards, e.g.: • Develop common architectural patterns,principles and standards to enable con-fident integration of autonomy technolo-gies; invest in updates to development andoperations processes to enable the trust-worthy deployment of increasingly au-tonomous missions • Update mission selection and review pro-cesses to consider assessment of agencyrisk posture, e.g., the risk of not includ-ing new technology or methods in NASAplanetary science missions. • With inter-directorate coordination, in-vest in laboratory and virtual autonomytestbeds so NASA centers and industrycan test and validate autonomy softwareand train on new processes. (See Rec • Spur the adoption of integrated au-tonomous functions in industry to supportNASA’s competed missions. (See Rec • Adopt new maturity evaluations for soft-ware and model trust to augment TRL inassessing autonomous functions, their in-tegration, and applications.4.
Set goals to reduce operations costsand determine the degree of au-tonomous operations required toachieve these goals.
NASA shouldcommission an independent study, poten-tially performed by the National Academyof Sciences, assessing existing operationsto evaluate how operations costs can bereduced by adopting autonomy and whatfunding and savings profiles would result.
Thanks to Ellen Van Wyk (NASA JPL) for illus-trations.
References [1] Stewart A. Collins.
The Mariner 6 and 7pictures of Mars . Scientific, Technical In-formation Office, National Aeronautics, andSpace Administration, 1971.72] Mark Brown.
Programmable Deep SpaceAutonomy: The First 25 Years . Apr. 1997.[3]
Planetary Mission Concept Studies Work-shop 2020 . . (Accessedon 09/08/2020).[4] . https : / / science . nasa .gov / technology / 2018 - autonomy -workshop/output- results . (Accessed on08/13/2020).[5] Carl Sagan and Rajashekar Reddy. Ma-chine Intelligence and Robotics Report ofthe NASA Study Group . 1980.[6] David J Evans. “OPS-SAT: FDIR Designon a Mission that Expects Bugs-and Lots ofThem”. In: . 2016, p. 2481.[7] NASA Autonomous Systems – SystemsCapability Leadership Team.
AutonomousSystems Taxonomy . 2019.[8] Steve Chien and Kiri L Wagstaff. “Roboticspace exploration agents”. In:
ScienceRobotics
Johns Hopkins APLTechnical Digest
Deep Impact Au-tonomous Navigation: the trials of targetingthe unknown . 2006.[11] Andres Castano et al. “Automatic detec-tion of dust devils and clouds on Mars”.In:
Machine Vision and Applications
ACM Transactions on Intelli-gent Systems and Technology (TIST)
Using autonomy flightsoftware to improve science return on EarthObserving One . 2005. [14] R Francis et al. “AEGIS autonomous tar-geting for ChemCam on Mars Science Lab-oratory: Deployment and results of initialscience team use”. In:
Science Robotics arXivpreprint arXiv:2007.15129 (2020).[16] Bethany Theiling et al. “Non-Robotic Sci-ence Autonomy Development”. In:
Plan-etary Science Decadal 2020 White Paper (2020).[17] Benjamin Donitz and et al.
New FrontiersMission Concept Study to Explore OortCloud Comets . 2020.[18] Shyam Bhaskaran et al. . 2018.[19] Kiri L Wagstaff et al. “Enabling OnboardDetection of Events of Scientific Interest forthe Europa Clipper Spacecraft”. In:
Pro-ceedings of the 25th ACM SIGKDD Inter-national Conference on Knowledge Discov-ery & Data Mining . 2019, pp. 2191–2201.[20] Glenn Reeves and et al. “Autonomy Chal-lenges and Development for Europa Lan-der”. In:
Planetary Decadal 2020 White Pa-per (2020).[21] D. Wang et al. “Using Flexible Execution,Replanning, and Model Parameter Updatesto Address Environmental Uncertainty fora Planetary Lander”. In:
Intl Symposiumon AI, Robotics, and Automation for Space (2020).[22] Martha Gilmore et al. “Venus Flagship Mis-sion Planetary Decadal Study”. In:
Interna-tional Planetary Probe Workshop WebinarSeries . 2020.[23] John Elliott, Mark Robinson, and etal. “Intrepid Planetary Mission ConceptOverview”. In:
International PlanetaryProbe Workshop . 2020, Webinar Session 6.[24] Carolyn Ernst and et al.
Mercury Lander . url : .825] Sanae Kubota, Carolyn Ernst, and et al.“Mercury Lander Planetary mission con-cept study for the 2023-2032 Decadal”.In: International Planetary Probe Work-shop Webinar Series . 2020.[26] Chad Edwards and et al. “Emerging Capa-bilities for Mars Exploration”. In:
PlanetaryScience Decadal 2020 White Paper (2020).[27] LH Matthies et al. “Robotics Technologyfor In Situ Mobility and Sampling”. In:
Planetary Science Decadal 2020 White Pa-per (2020).[28] Daniel Gaines et al.
Self-Reliant Rover De-sign for Increasing Mission Productivity .2018.[29] Soumyo Dutta and et al. “Aerocapture asan Option for Ice Giants Missions”. In:
In-ternational Planetary Probe Workshop We-binar Series . 2020.[30] Alex Austin and et al. “Enabling and En-hancing Science Exploration Across the So-lar System: Aerocapture Technology forSmallSat to Flagship Missions”. In:
Inter-national Planetary Probe Workshop Webi-nar Series . 2020.[31] T. Marshall Eubanks et al. “Exobodies inOur Back Yard: Science from Missions toNearby Interstellar Objects”. In:
PlanetaryDecadal 2020 White Paper (2020).[32] Julie Castillo and John Brophy.
Ceres Plan-etary Mission Concept Study: SamplingBrines from an Evolved Ocean World . 2020.[33] Wendy Calvin.
PCMS MORIE. A Plan-etary Mission Concept Study: Mars Or-biters for Resources, Ices and Environ-ments . 2020.[34] Rob Lillis and the MOSAIC Team.
MO-SAIC: A Planetary Mission Concept Study.Mars Orbiters for Surface-Atmospheric-Ionospheric Connections . 2020.[35] Carly Howett.
Persephone: A Pluto-SystemOrbiter & Kupier Belt Explorer . 2020.[36] Daniel Grebow et al. “Dawn safing ap-proach to Ceres re-design”. In:
AIAA/AASAstrodynamics Specialist Conference . 2016,p. 5426. [37] Ian Garrick-Bethell. “NanoSWARM: ANano-satellite Mission to Measure Particlesand Fields Around the Moon”. In:
AGUFM
Today’s Deep Space Com-munications System: The Earth-based Ele-ments . 2019.[39] Lori Sherea Glaze and James Brian Garvin.“VICI (Venus in situ composition inves-tigations): the next step in understand-ing Venus climate evolution”. In:
AGUFM
Surface and atmosphere geo-chemical explorer (SAGE) baseline designfrom March 2003 Team X studies . 2003.[41] The Planetary Society.