Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 1 Dr. Werner J.A. Dahm Director, Security & Defense Systems Ini3a3ve Arizona State University Perspec1ves on Verifica1on and Valida1on in Complex Adap1ve Systems Developing ways to establish “certifiable trust” in emerging highly adaptable autonomous systems Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 1
Dr. Werner J.A. Dahm Director, Security & Defense Systems Ini3a3ve Arizona State University
Perspec1ves on Verifica1on and Valida1on in Complex Adap1ve Systems
Developing ways to establish “certifiable trust” in emerging highly adaptable autonomous systems
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 2
“Simple” Complex Adaptive Systems
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 3
“Complex” Complex Adaptive Systems
n Premium cars today are controlled by O(100) interconnected ECUs
n E.g., autobraking, lane drift warning, autonomous parallel parking, etc.
n 70% of development costs and 40% of production costs are software systems
n Systems today are highly federated but subsystem interactions are increasing
n 5th-generation fighter A/C are deeply integrated software-driven systems
n Overwhelming majority of functions are driven autonomously by software
n Software development and validation account for over 50% of costs
n DoD systems (adaptive radars, UCAVs, etc.) are now increasingly autonomous
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 4
Elements of Complex Adaptive Systems
n Systems composed of many interacting agents, with system evolving over time via adaptation n Agents are at least partly autonomous n No agent has full global view of the system n Decentralized structure in which no agent
has full control of the others
n In some systems all agents may be identical and simple; in others all are different and non-simple
n Resulting complex system behavior is more than ‘sum of the parts’ – nonlinear dynamics
n Emergent high-level system behavior driven by dynamics of simultaneous interactions of agents
n Typically cannot be fully inferred or anticipated from behaviors of the individual agents
n Number of agents and nature of interactions matter; threshold for emergent behavior
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 5
Enablers of Complex Adaptive Systems
n Miniaturization n MEMS fabrication
n Ubiquitous sensing n Small, cheap, reliable n Multi-modality
n Embedded processing n Multi-core, GPGPUs n Processing speeds
n Storage n Compact terabyte-scale n Inexpensive, solid-state
n Software adaptation/learning n Expert systems n Genetic algorithms n Machine learning n Stochastic dither
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 6
U.S. Air Force “Technology Horizons”
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 7
Highly Adaptable Autonomous Systems
n Capability increases, manpower efficiencies, and cost reductions are possible through far greater use of autonomous systems
n Dramatic in degree of autonomy and range of systems and processes where autonomous reasoning and control can be applied
n Adaptive autonomy can offer time-domain operational advantages over adversaries using human planning and decision loops
n S&T to establish “certifiable” trust in highly adaptible autonomous systems is a key to enabling this transformation
n Potential adversaries may gain benefits from fielding such systems without any burden of establishing certifiable “trust in autonomy”
n As one of the greatest beneficiaries of such autonomous systems, the Air Force must lead in developing the underlying S&T basis
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 8
Highly Adaptable Autonomous Systems
AT&T Global Network Operations Center
FAA NextGen ATC Smart Grid
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 9
Highly Adaptable Autonomous Systems
n UAS moving beyond traditional surveillance and kinetic strike roles
n Longer-endurance missions require high-efficiency engine technologies
n In-flight automated refueling will be key for expanding UAS capabilities
n May include ISR functions beyond traditional electro-optic surveillance
n LO may allow ops in contested or denied (non-permissive) areas
n Electronic warfare (EW) by stand-in jamming is a possible future role
n Wide-area airborne surveillance (WAAS) is increasingly important
n Directed energy strike capability is likely to grow (laser and HPM)
n Civil uses include border patrol and interdiction, and humanitarian relief
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 10
UAS Automated Aerial Refueling (AAR)
n Aerial refueling of UAVs from USAF tanker fleet is essential for increasing range and endurance
n Requires location sensing and relative navigation to approach, hold, and move into fueling position
n Precision GPS can be employed to obtain needed positional information
n Once UAV has autonomously flown into contact position, boom operator engages as normal
n Key issues include position-keeping with possible GPS obscuration by tanker and gust/wake stability
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 11
Flight Testing of UAS AAR Algorithms
n August 2006 initial flight tests of AFRL-developed control algorithms for automated aerial refueling
n KC-135 with Learjet-surrogate UAS platform gave first “hands-off” approach to contact position
n Subsequent positions and pathways flight test and four-ship CONOPS simulations successful
n 120 mins continuous “hands-off” station keeping in contact position; approach from ½-mile away
n 12 hrs of “hands-off” formation flight with tanker including autonomous position-holding in turns
n Position-holding better than human-piloted flight
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 12
Autonomous Sense-and-Avoid (SAA)
n Sense-and-Avoid was Global Hawk ATD for in-flight collision avoidance system
n Flight on surrogate aircraft began 2006
n Autonomous detection and avoidance of cooperative & non-cooperative intruders
n Jointly Optimal Collision Avoidance (JOCA) was transition program in 2009
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 13
UAS: Expanding Roles/New Concepts
n New unmanned aircraft systems (VULTURE) and airships (ISIS) can remain aloft for years
n Delicate lightweight structures can survive low-altitude winds if launch can be chosen
n Enabled by solar cells powering lightweight batteries or regenerative fuel cell systems
n Large airships containing football field size radars give extreme resolution/persistence
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 14
Distributed/Cooperative UAS Control
n Task coupling of multiple UAVs is key for many missions; e.g. urban areas
n Allows multiple UAVs to act as single coordinated unit to meet mission need
n Allow dynamic task re-assignment to reduce overall operator workload
n Demonstrated in Talisman Saber 2009
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 15
Integration of Complex Adaptive Systems
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 16
DoD S&T Priorities for FY 13-17
1. Data to Decisions 2. Engineered Resilient Systems 3. Cyber Science and Technology 4. Electronic Warfare / Electronic Protection 5. Counter Weapons of Mass Destruction 6. Autonomy 7. Human Systems
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 17
ASD(R&E) Systems 2020 Initiative
Systems 2020 is an Initiative from the Assistant Secretary of Defense
(Research & Engineering) $100M over 5 years
(FY12 – 16)
Basic Science Issues are indentified for each of these areas
1. Model-Based Engineering • M&S throughout the development process • More effective concept engineering • Concurrent design, develop, deploy, and evolve
2. Platform-Based Engineering 3. Capability on Demand
• Embedded organic adaptation capabilities • On-demand adaptability • Self-adaptive systems • Field-adaptive systems
4. Trusted Systems Design
Systems 2020 Research Areas
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 18
Adaptable Autonomous System Attributes
Four Required Attributes of Highly Adaptable Autonomous Systems
n Large number of on/off-board inputs (embedded sensors; other platforms)
n Large number of interdependent decisions (via fusion of input data)
n Large number of system adaptations (may be deterministic or stochastic)
n Adaptations evolve continuously based on feedback-driven learning
As the number of inputs and the number of possible system adaptations increase, these become near-infinite state systems
Mostly driven by number of adaptations
Makes it hugely difficult to prove that such systems will always operate as intended
Workshop on Verification and Validation in Computational Science, Notre Dame University, 17-19 October 2011 19
Software Growth & Error Rates/Types
System ESLOC F-4A 1,000
F-15A 50,000
F-16C 300,000
F-22 2,500,000
Android OS 12,000,000
F-35A 18,000,000
2011 BMW 535i >80,000,000
Simple CAS
Adaptability None User Adaptive Competitor Adaptive