Big Picture for Autonomy Research in DoD

Transcription

1 Big Picture for Autonomy Research in DoD Approved for Public Release Soft and Secure Systems and Software Symposium Dr. Robert Grabowski Jun 9, 2015 For internal MITRE use

2 2 Robotic Experience 9 Years US Navy nuclear power program PhD CMU, small heterogeneous teams 15 years of robotics research 2005 DARPA Grand Challenge finalist Novel perception Test and evaluation, metrics USS Truxtun Physics CGN-35 Millibots Meteor Centaurs

3 3 Agenda Tenets of autonomy research Autonomy survey evolution Definitions and spaces Survey of autonomy in DoD Autonomy in motion Acceptance gap

4 4 Tenets of Autonomy Research Focus on uniqueness Where autonomy is unique and how this uniqueness impacts the tester Why current approaches insufficient, what new approaches are needed Recognize perspectives Understand how each looks at autonomy different, what drives them Develop an appropriate language to communicate across perspectives Pivot from development of autonomy to test of autonomy Drive with models Provide relationships between perspectives Assure completeness / gap analysis (don t leave holes) Mechanism for decomposition (towards a plan) Make it actionable Use insights to drive investment, new technologies, methodologies Build in capability to measure progress

5 5 Autonomy History, definitions and fights

6 6 Evolution of thinking Early investigation focused on platforms, mostly provided portfolio and catalog 2010

7 7 Evolution of thinking Autonomy becomes S&T priority SecDef RDE Focus

8 8 Evolution of thinking Autonomy becomes S&T priority First systematic focus on autonomy Introduction of MUM-T Introduction of issue of trust SecDef RDE Focus 2011 DoD Integrated Roadmap

9 9 Evolution of thinking Autonomy becomes S&T priority First systematic focus on autonomy Introduction of MUM-T Introduction of issue of trust Shift from platforms to decisions Shift towards science of autonomy SecDef RDE Focus 2011 DoD Integrated Roadmap 2012 DSB Role Autonomy

10 10 Evolution of thinking Autonomy becomes S&T priority First systematic focus on autonomy Introduction of MUM-T Introduction of issue of trust Shift from platforms to decisions Shift towards science of autonomy Emphasis on calibrated trust Awareness of testing and V&V SecDef RDE Focus 2011 DoD Integrated Roadmap 2013: DoD Autonomous Research Initiative 2012 DSB Role Autonomy

11 11 Evolution of thinking Autonomy becomes S&T priority First systematic focus on autonomy Introduction of MUM-T Introduction of issue of trust Shift from platforms to decisions Shift towards science of autonomy Emphasis on calibrated trust Awareness of testing and V&V SecDef RDE Focus 2011 DoD Integrated Roadmap Emphasis and constraints on autonomous weaponization 2013: DoD Autonomous Research Initiative 2012 DSB Role Autonomy 2013: SecDef Autonomous Weaponization

12 12 Evolution of thinking Autonomy becomes S&T priority First systematic focus on autonomy Introduction of MUM-T Introduction of issue of trust Shift from platforms to decisions Shift towards science of autonomy Emphasis on calibrated trust Awareness of testing and V&V SecDef RDE Focus 2011 DoD Integrated Roadmap Emphasis and constraints on autonomous weaponization Focus on autonomy as decision Establishment of COI 2013: DoD Autonomous Research Initiative 2012 DSB Role Autonomy 2013: SecDef Autonomous Weaponization 2014: ASDRE Reliance 21

13 13 Evolution of thinking Autonomy becomes S&T priority First systematic focus on autonomy Introduction of MUM-T Introduction of issue of trust Shift from platforms to decisions Shift towards science of autonomy Emphasis on calibrated trust Awareness of testing and V&V SecDef RDE Focus 2011 DoD Integrated Roadmap Emphasis and constraints on autonomous weaponization Focus on autonomy as decision Establishment of COI 2013: DoD Autonomous Research Initiative 2012 DSB Role Autonomy Emphasis on HRI, Teams Emphasis on T&E/V&V Technology Offset 2013: SecDef Autonomous Weaponization 2014: ASDRE Reliance : Autonomy Focus

14 14 Tenets of Autonomy Research Focus on uniqueness Where autonomy is unique and how this uniqueness impacts the tester Why current approaches insufficient, what new approaches are needed Recognize perspectives Understand how each looks at autonomy different, what drives them Develop an appropriate language to communicate across perspectives Pivot from development of autonomy to test of autonomy Drive with models Provide relationships between perspectives Assure completeness / gap analysis (don t leave holes) Mechanism for decomposition (towards a plan) Make it actionable Use insights to drive investment, new technologies, methodologies Build in capability to measure progress

15 15 Autonomy Definitions Many definitions, many opinions, many arguments All good but somehow all equally unsatisfying

16 16 Autonomy Definitions Many definitions, many opinions, many arguments All good but somehow all equally unsatisfying Webster Dictionary (2015) the state of existing or acting separately from others AFRL Autonomy Science and Technology Strategy (2014) systems which have a set of intelligence-based capabilities that allow it to respond to situations that were not pre-programmed or anticipated in the design. Autonomous systems have a degree of self-government and self-directed behavior Towards Safety Assurance of Trusted Autonomy (2012) the ability to reason and make decisions to reach given goals based on a systems current knowledge and its perception of the variable environment in which it evolves DoD Unmanned Systems Integrated Roadmap (2011) autonomous systems develop, for themselves, the laws and strategies by which they choose their behavior. Army GO Workshop (2009) a system which, by its appearance or movements, conveys a sense that it has intent or agency of its own NIST Autonomy Levels for Unmanned Systems Framework (2004) the condition or quality of being self-governing

17 17 Autonomy Definitions Many definitions, many opinions, many arguments All good but somehow all equally unsatisfying Webster Dictionary (2015) the state of existing or acting separately from others AFRL Autonomy Science and Technology Strategy (2014) systems which have a set of intelligence-based capabilities that allow it to respond to situations that were not pre-programmed or anticipated in the design. Autonomous systems have a degree of self-government and self-directed behavior Towards Safety Assurance of Trusted Autonomy (2012) the ability to reason and make decisions to reach given goals based on a systems current knowledge and its perception of the variable environment in which it evolves DoD Unmanned Systems Integrated Roadmap (2011) autonomous systems develop, for themselves, the laws and strategies by which they choose their behavior. Army GO Workshop (2009) a system which, by its appearance or movements, conveys a sense that it has intent or agency of its own NIST Autonomy Levels for Unmanned Systems Framework (2004) the condition or quality of being self-governing

18 18 Autonomy Definitions Many definitions, many opinions, many arguments All good but somehow all equally unsatisfying Webster Dictionary (2015) the state of existing or acting separately from others AFRL Autonomy Science and Technology Strategy (2014) systems which have a set of intelligence-based capabilities that allow it to respond to situations that were not pre-programmed or anticipated in the design. Autonomous systems have a degree of self-government and self-directed behavior Towards Safety Assurance of Trusted Autonomy (2012) the ability to reason and make decisions to reach given goals based on a systems current knowledge and its perception of the variable environment in which it evolves DoD Unmanned Systems Integrated Roadmap (2011) autonomous systems develop, for themselves, the laws and strategies by which they choose their behavior. Army GO Workshop (2009) a system which, by its appearance or movements, conveys a sense that it has intent or agency of its own NIST Autonomy Levels for Unmanned Systems Framework (2004) the condition or quality of being self-governing I know it when I see it Justice Potter Stewart 1964

19 19 Tenets of Autonomy Research Focus on uniqueness Where autonomy is unique and how this uniqueness impacts the tester Why current approaches insufficient, what new approaches are needed Recognize perspectives Understand how each looks at autonomy different, what drives them Develop an appropriate language to communicate across perspectives Pivot from development of autonomy to test of autonomy Drive with models Provide relationships between perspectives Assure completeness / gap analysis (don t leave holes) Mechanism for decomposition (towards a plan) Make it actionable Use insights to drive investment, new technologies, methodologies Build in capability to measure progress

20 20 Autonomy Levels Parallel approach is to divide autonomy into levels Shift from singularity to spectrum to spaces NASA (Function Specific Level of Autonomy) Bloom Levels of Autonomy DoD 4 levels of Autonomy (USIR) Flavors of Autonomy Army (Think, Look, Move, Talk, Work) SAE Levels of Driving Autonomy NIST Autonomy Space Nice for categorization but usefulness still in question

21 21 Tenets of Autonomy Research Focus on uniqueness Where autonomy is unique and how this uniqueness impacts the tester Why current approaches insufficient, what new approaches are needed Recognize perspectives Understand how each looks at autonomy different, what drives them Develop an appropriate language to communicate across perspectives Pivot from development of autonomy to test of autonomy Drive with models Provide relationships between perspectives Assure completeness / gap analysis (don t leave holes) Mechanism for decomposition (towards a plan) Make it actionable Use insights to drive investment, new technologies, methodologies Build in capability to measure progress

22 22 Autonomy Perspectives Autonomy assessment Listen to what agencies are saying Look at what they think autonomy will enable No single group captures autonomy Collectively they build a full picture 30+ DoD, Service roadmaps, strategic plans and investment strategies referencing autonomy Autonomy Survey Paper

23 23 High Level Review of 100+ documents DoD, Agencies Federal and Civil, International and Academic 30+ critical documents identified 2015 OTI Technical Assessment Autonomy 2015 ATEVV Investment Strategy 2015 Unmanned Autonomous Systems Test Strategy 2015 Naval S&T Strategy 2014 AFOSR Technical Strategic Plan 2013 DoD Unmanned Systems Integrated Roadmap 2012 DSB The Role of Autonomy in DoD Systems 2012 DoD Autonomy in Weapon Systems Directive 2012 DoD Autonomy Research Pilot Initiative 2011 RSJPO Unmanned Ground Systems Roadmap 2011 DoD Unmanned Systems Integrated Roadmap 2011 DoD Science and Technology Priorities for Fiscal Years Planning 2010 NASA, Robotics, Tele-Robotics and Autonomous Systems Roadmap 2010 Army Unmanned Air Systems Roadmap 2010 Air Force Technology Horizons 2009 DARPA Strategic Plan 2009 ARCIC Robotics Strategic White Paper 2009 Air Force Unmanned Aircraft Systems Flight Plan 2009 DoD Unmanned Systems Integrated Roadmap 2007 Navy Unmanned Surface Vehicle Master Plan

24 24 Dual perspective Anticipation A closer look at what stakeholders are anticipating from autonomy Look towards new capabilities enabled by advances in technology and potentially new applications and approaches Implicitly speaks to new challenges as well Anticipation Trepidation Trepidation The opposite perspective capturing the concerns with adoption of unstructured autonomy Tends to be on concerns of consequences, vulnerabilities and liabilities Both sides essential in true understanding of autonomy

25 25 Anticipations Reduction of Manpower - realizing the promise of unmanned systems to reduce manpower and cost Mitigation of unmanned reduce manpower, cost, logistics of existing platforms Reduction of operators further reduction of manning and specially qualified operators to control more than one platform or asset Information filtering reduction of sheer data volume collected by unmanned systems. Systems that make decisions on what not to show

26 26 Anticipations Reduction of Manpower - realizing the promise of unmanned systems to reduce manpower and cost Mitigation of unmanned reduce manpower, cost, logistics of existing platforms Reduction of operators further reduction of manning and specially qualified operators to control more than one platform or asset Information filtering reduction of sheer data volume collected by unmanned systems. Systems that make decisions on what not to show Tactical Advantage - added advantages on the battlefield Faster reaction time local decisions faster than human cycle Deeper penetration operation in inaccessible or denied environments Extended operation can operate longer than human cycles Agility and adaptation ability to adjust to changing environment, changing mission goals, ability to use in secondary missions

27 27 Anticipations Reduction of Manpower - realizing the promise of unmanned systems to reduce manpower and cost Mitigation of unmanned reduce manpower, cost, logistics of existing platforms Reduction of operators further reduction of manning and specially qualified operators to control more than one platform or asset Information filtering reduction of sheer data volume collected by unmanned systems. Systems that make decisions on what not to show Tactical Advantage - added advantages on the battlefield Faster reaction time local decisions faster than human cycle Deeper penetration operation in inaccessible or denied environments Extended operation can operate longer than human cycles Agility and adaptation ability to adjust to changing environment, changing mission goals, ability to use in secondary missions Trusted Companion - System capable of providing real-time, tactical and proximate support to warfighters Faithful servant utilization of competent mules, closer proximity to humans, operations not in contact with adversary Loyal wingman high tempo coordination and interaction, operations in contact with adversary

28 28 Trepidations Unmanaged Complexity - Complexity exceeds ability to design, monitor, and test Coupling of sub-systems inability to understand interactions between systems, complexity of test, difficulty in prediction Unstructured adaptation implications of systems that can learn new rules Brittleness superior behavior in ideal conditions, potentially catastrophic failure (or behavior outside nominal

29 29 Trepidations Unmanaged Complexity - Complexity exceeds ability to design, monitor, and test Coupling of sub-systems inability to understand interactions between systems, complexity of test, difficulty in prediction Unstructured adaptation implications of systems that can learn new rules Brittleness superior behavior in ideal conditions, potentially catastrophic failure (or behavior outside nominal Unintended Consequences difficulty in understanding true risks to mission or force structure. Ambiguity of command Complications on command and command structure Safety and liability political and legal fallout from accident or damage Ethics of weaponization implications on rules of engagement, long term complications (consider current issue with drones) Complacency difficulty in keeping human awareness at correct level of attention

30 30 Trepidations Unmanaged Complexity - Complexity exceeds ability to design, monitor, and test Coupling of sub-systems inability to understand interactions between systems, complexity of test, difficulty in prediction Unstructured adaptation implications of systems that can learn new rules Brittleness superior behavior in ideal conditions, potentially catastrophic failure (or behavior outside nominal Unintended Consequences difficulty in understanding true risks to mission or force structure. Ambiguity of command Complications on command and command structure Safety and liability political and legal fallout from accident or damage Ethics of weaponization implications on rules of engagement, long term complications (consider current issue with drones) Complacency difficulty in keeping human awareness at correct level of attention Smart Adversary impact from contact with an interactive and aware opponent Corruption impacts of compromised systems including ability to detect Loss of control adversary gaining control of asset or information The aware adversary how engagement change when adversary aware they are interacting with artificial system Asymmetry of development concern of advancements of technology from an adversary without the political, legal or self-imposed constraints we may be under

31 31 Tenets of Autonomy Research Focus on uniqueness Where autonomy is unique and how this uniqueness impacts the tester Why current approaches insufficient, what new approaches are needed Recognize perspectives Understand how each looks at autonomy different, what drives them Develop an appropriate language to communicate across perspectives Pivot from development of autonomy to test of autonomy Drive with models Provide relationships between perspectives Assure completeness / gap analysis (don t leave holes) Mechanism for decomposition (towards a plan) Make it actionable Use insights to drive investment, new technologies, methodologies Build in capability to measure progress

32 32 DSB Recommendations The DoD should abandon the debate over definitions [and] of levels of autonomy Focus on how autonomy impacts things not what it means

33 33 DSB Recommendations The DoD should abandon the debate over definitions [and] of levels of autonomy Focus on how autonomy impacts things not what it means The Military Services should structure autonomous systems acquisition programs to separate the autonomy software from the vehicle platform. Autonomy is about decisions not platforms

34 34 DSB Recommendations The DoD should abandon the debate over definitions [and] of levels of autonomy Focus on how autonomy impacts things not what it means The Military Services should structure autonomous systems acquisition programs to separate the autonomy software from the vehicle platform. Autonomy is about decisions not platforms AT&L should create developmental and operational test and evaluation techniques that focus on the unique challenges of autonomy Focus on unique properties of autonomy Focus on testing and role in adoption

35 35 Autonomy in Motion Acceptance Gap

36 36 Autonomy Spectrum Autonomy Spectrum scripted semantic responsive adaptive self aware Autonomy not sole purview of unmanned

37 37 Autonomy at Rest Autonomy at Rest degree of autonomy scripted self aware Siri IBM Watson Distributed Battle Mgt High Frequency Trading Next Gen ATC Autonomy at rest Cognitive assistants (siri) Expert systems (high frequency trading) Expert systems (watson) C4I (next gen air traffic control) C4I (Distributed battle management, Rspace )

38 38 Autonomy Acceptance Autonomy at Rest scripted self aware Acceptance Threshold IBM Watson For any system, have some notional concept of acceptance Based on nature of autonomous system and environment it is in

39 39 Autonomy Acceptance Autonomy at Rest Constrain application scripted self aware Increase oversight, introspection For any system, have some notional concept of acceptance Based on nature of autonomous system and environment it is in Adjustments Can constrain that autonomy Can mitigate with greater oversight IBM Watson Can develop new concepts for introspection and acceptance

40 40 Autonomy Acceptance Autonomy at Rest scripted Acceptance Threshold self aware IBM Watson Result is a system that can be released in the wild and vetted through employment Evolutionary approach to acceptance Adjust autonomy to match acceptance criterion Develop confidence over time Understand nature of application of autonomy

41 41 Acceptance/Trust Gap Autonomy in Motion scripted self aware Squad Support Robot Convoy Cargo Lift Civilian Cargo ASW warfare Driverless Taxi Long duration ISR Fire fighting Wingman Unmanned autonomous systems (UAS) present a unique challenge to the autonomy world Autonomy may require a minimum threshold just to operate

42 42 Acceptance/Trust Gap Autonomy in Motion scripted self aware Squad Support Robot Convoy Cargo Lift Civilian Cargo ASW warfare Driverless Taxi Long duration ISR Fire fighting Wingman Some applications may be mitigated with oversight, co-human control Concept breaks when you remove the man (cant have a ½ autonomous system)

43 43 Acceptance/Trust Gap Autonomy in Motion scripted Acceptance Threshold self aware Cargo Lift Consider Autonomous air lift (like AACUS) Autonomous takeoff and landing Landing zone assessment and selection Obstacle avoidance and negotiations (powerlines, trees) Payload negotiation and management Environmental adjustment (weather, wind, updraft) Adversarial interaction

44 44 Acceptance/Trust Gap Autonomy in Motion scripted Acceptance Threshold self aware Cargo Lift Can restrict some of the operations but ultimately gap still remains

45 45 Acceptance/Trust Gap Autonomy in Motion GAP scripted Acceptance Threshold self aware Cargo Lift Result is an acceptance/trust gap that restricts release and operations Restricted release retards co-evolution Not just about making a more competent system, need to overcome trust gap

46 46 Acceptance/Trust Gap ATEVV Investment Strategy Automated requirement (codable language) Traceable evidence (pedigree) Progressive sequential M&S, T&E (prediction) Real time monitoring (watchdog oversight) Just-in-time prediction (behaviors, actions) Reusable assurance (actuaries) ATEVV Cargo Lift Unmanned Autonomous Systems Test Targeted testing (salience) Smart test range (adaptive to system) HMI efficacy (distributed SA) Decision transparency (rules and impact) Two main groups focusing on test technologies ATEVV under DoD ASDR&E UAST under DoD TRMC UAST

47 47 Tenets of Autonomy Research Focus on uniqueness Where autonomy is unique and how this uniqueness impacts the tester Why current approaches insufficient, what new approaches are needed Recognize perspectives Understand how each looks at autonomy different, what drives them Develop an appropriate language to communicate across perspectives Pivot from development of autonomy to test of autonomy Drive with models Provide relationships between perspectives Assure completeness / gap analysis (don t leave holes) Mechanism for decomposition (towards a plan) Make it actionable Use insights to drive investment, new technologies, methodologies Build in capability to measure progress

48 Autonomy: A decade of understanding Discussion 48