afrl autonomy -...
TRANSCRIPT
1
Integrity Service Excellence
AFRL Autonomy
11 July 2013
Dr. Jim Overholt AFRL Senior Scientist for Autonomy
Human Effectiveness Directorate (711HPW/RH) Air Force Research Laboratory
Kris Kearns AFRL Autonomy Portfolio Lead
Human Effectiveness Directorate (711HPW/RH) Air Force Research Laboratory
Distribution A. Approved for public release: distribution unlimited. (88ABW-2013-3169, 10 July 2013)
2
Topics
AFRL Autonomy
Autonomy Challenges in Air Domain
DoD Autonomy Priority Steering Council (PSC) & Community of Interest (COI)
3
Not really!
in Name Only. . .
Unmanned Air Vehicle
Unmanned Air Vehicle Leadership
Admin & Overhead
Pilots Sensors Ops
Maint Mission Coord
Processing, Exploitation, Dissemination (PED)
Full Motion Video
Signal Intelligence
Maint
Pilots
Sensor Ops
Maintenance
The Automation-Autonomy Continuum
The system functions with no/little human operator involvement, well-defined tasks that have predetermined responses
Automation Systems have intelligence-based capabilities, respond to situations not pre-programmed or anticipated in design
Autonomy
Autonomy is a capability that enables a particular action of a system to be automatic or, within programmed boundaries, “self-governing.”
(The Role of Autonomy in DoD Systems, Defense Science Board, July 2012)
― Manpower efficiencies ― Rapid response and 24/7 presence ― Harsh environments ― New mission requirements ― Capabilities beyond human limits …….. Across Operational Domains
Decentralization, Uncertainly, Complexity…Military Power in the 21st Century will be defined by our ability to adapt – adaptation is THE underlying foundation of
autonomous technology
Key DoD Challenges Addressed by Autonomy
Some Shared Autonomy S&T Challenges for DoD & Industry Partners
Human/Autonomous System Interaction and Collaboration
Test, Evaluation, Validation, and Verification
Scalable teaming of Autonomous Systems
Machine Perception, Reasoning and Intelligence
Future R&D must provide secure communication between agents and their operators, expand shared perception and problem solving across multiple agents, and advance guidance/control
Future R&D must further integrate artificial intelligence & human cognitive models, advance human-agent feedback loops, optimize trust/transparency, and advance sensor/data decision models
Future R&D must advance data-driven analytics, contingency-based control strategies, decision making algorithms to enable operations, and adaptive guidance & control
Must expand its TEV&V capabilities in live and simulated environments across all operational domains. Test beds must incorporate scenario-based testing.
7
Topics
AFRL Autonomy
Autonomy Challenges in Air Domain
DoD Autonomy Priority Steering Council (PSC) & Community of Interest (COI)
8
AFRL Autonomy Team
• RH – Mike Patzek, RHCI* – Mark Draper, RHCI – Scott Galster, RHCP – Jeff Graley, RHXM
• RI – Jerry Dussault, RISB – Rick Metzger, RIS*
• RQ – Bob Smith, RQCC** – Corey Schumacher, RQCA* – Jake Hinchman, RQCC
• RV – Scott Erwin, RVSV – Paul Zetocha, RVSV* – Khanh Pham, RVSV
• RW – Rob Murphey, RWW – Will Curtis, RWWN* – TJ Klausutis, RWW – Ric Wehling, RWWI
• RY – Raj Malhotra, RYAR*
• OSR – Tristan Nguyen, RTC
Autonomy Research conducted at many of the AFRL Technical Directorates
9
Autonomous Systems and Technologies Cut Across Domains
Cyber – systems handle massive, distributed, and data/information-intensive tasks
Aircraft – systems operate in complex environment needing to synchronize space and mission mgmt
Weapons – systems that coordinate mission execution
Space – once launched systems must operate “on their own” in a harsh environment
10 Ensure safe and effective systems in unanticipated & dynamic environments
AFRL Autonomy Vision & Goals
Intelligent machines seamlessly integrated with humans - maximizing mission performance in
complex and contested environments
Create actively coordinated teams of multiple machines
Ensure operations in complex, contested environment
Demonstrate highly effective human-machine teaming
11
Ensure safe and effective systems in unanticipated & dynamic environments
AFRL Autonomy Goal Human-Machine Teaming
Create actively coordinated teams of multiple machines
Ensure operations in complex, contested environment
Intelligent machines seamlessly integrated with humans - maximizing mission performance in
complex and contested environments
Demonstrate highly effective human-machine teaming
12
Ensure systems are safe and effective in a wide variety of unanticipated and rapidly evolving environments
AFRL Autonomy Human-Machine Teaming
Intelligent machines seamlessly integrated with humans - maximizing mission performance in
complex and contested environments
Create actively coordinated teams of multiple machines
Ensure operations in complex, contested environment
Demonstrate highly effective human-machine teaming Transparency
Communication
Training Sensing
• Enable & Calibrate trust between human and machines
• Develop common understanding and shared perception between humans and machines
• Create an environment for flexible and effective decision making
Interfaces
13
SENSE
ASSESS AUGMENT
attached to skin
detached f rom skin
5mm
Human-Machine Teaming Augmentation Framework
•Extraction of objective human measures to inform empirical studies
•Implement a controlled feedback cycle
14
Human State Measurement
• Developing measurement techniques for stress, workload, attention.
• Correlating human cognitive tasks to performance
• Long Term vision: Providing the machine data about the human’s state so the machine can aid mission performance
Human State Sensing foundational for humans and machines to work as a
team
15
Advanced Interfaces for Multi-UAV Control
Developing interface technologies to optimize human performance (increased decision-making, decreased stress, etc)
16
Balancing Operator Involvement w/ Automation
Developing Human-in-the-Loop Testbeds Objectively and subjectively measure human
performance Physiological (Eye tracking, ECG, voice analysis) Subjective (Situational Awareness, Trust, Usability) Mission Performance
… to ensure an optimized human-machine team
MULTI-UAV TESTBED ANALYST TESTBED
17
Ensure systems are safe and effective in a wide variety of unanticipated and rapidly evolving environments
AFRL Autonomy Goals and Major Objectives
Intelligent machines seamlessly integrated with humans - maximizing mission performance in
complex and contested environments
Ensure operations in complex, contested environment
Demonstrate highly effective human-machine teaming
• Mature machine Intelligence
• Develop and manage fractionated and composable systems
• Develop reliable, secure, interoperable communication
Create actively coordinated teams of multiple machines
Communication
Perceive Reasoning
Training
Sense
Act
Plan
18
Machine Self-Awareness: Adaptation to Degradations in Systems Health
Autonomous Systems need to be responsive to systems health • Determination of failure, or impending failure • Reconfiguration of control to allow for safe recovery, or • Adaptation to enable continued mission operations
Hierarchical health diagnosis architecture with feedback and reasoning for
disambiguation
Adaptive inner-loop (stability) and outer-loop (trajectory) control to recover from failures
System Health Reasoner
Subsystem State
Assessment
Subsystem State
Assessment
Subsystem State
Assessment
19
UAS Autonomy & Teaming
UAS Autonomy & Teaming: Key Goals
• Expand the available action space and decision space
• Operate in contested and denied environments • Increase coordination between assets • React faster than the opponents decision cycle
• Develop and demonstrate the control and autonomy technologies required to enable robust, adaptive, and coordinated combat operations by heterogeneous, mixed teams of air assets
• Cooperative ISR challenge is to provide ISR as an off-board “service” without the need to directly control the UAS team
• Future is Tactical Battle Manager (TBM) for multi-vehicle combat operations, supporting team mission execution in contested and denied environments
20
Ensure systems are safe and effective in a wide variety of unanticipated and rapidly evolving environments
AFRL Autonomy Goals and Major Objectives
Intelligent machines seamlessly integrated with humans - maximizing mission performance in
complex and contested environments
Create actively coordinated teams of multiple machines
Demonstrate highly effective human-machine teaming
• Develop technologies that assure robust system and self-protection capabilities
• Develop technologies that enable situational understanding of the contested environment
Ensure operations in complex, contested environment
21
Key enabling technology for integrating unmanned air vehicles into manned airspace – provides automated
“see & avoid” required of piloted aircraft
Deconfliction/Conflict Avoidance
• Demo’ed single-intruder autonomous detect/avoid using EO/TCAS (Dec 06) • Demo’ed single-, two-intruder closed-loop detect/avoid using EO/TCAS/ADS-
B/surrogate radar (Aug & Sep 09) • Final demo of single-, two-intruder closed loop detect/avoid with improved EO,
SAA radar, ADS-B, and TCAS
Sense & Avoid (SAA)
22
UAS Airspace Integration
UAS Airspace Integration: Capability Progression
Sensor, vehicle control algorithms, and pilot interface development and flight test
Common Airborne Sense and Avoid system, scalable to Group 3-5
Terminal Area Operations for safe, efficient ground and terminal operations
Onboard sensors such as radar, EO/IR, TCAS, and ADS-B will enable detection of both cooperative and non-cooperative aircraft, providing protection in all classes of airspace.
The ABSAA system will provide autonomous maneuvering or Pilot-In/On-The-Loop capability as operations dictate.
Key to success is exhibiting pilot-like behavior that allows seamless integration into normal flight operations
23
Ensure systems are safe and effective in a wide variety of unanticipated and rapidly evolving environments
AFRL Autonomy Goals and Major Objectives
• Provide assurance for machine intelligence and decision-making in complex, uncertain, and dynamic environments
• Develop methods to ensure reliability of human-machine communication and interaction
•Develop rigorous and verifiable architecture systems for data centric autonomous systems
•Develop methodology to V&V fractionated and composable systems
Intelligent machines seamlessly integrated with humans - maximizing mission performance in
complex and contested environments
Create actively coordinated teams of multiple machines
Ensure operations in complex, contested environment
Demonstrate highly effective human-machine teaming
Ensure safe and effective systems in unanticipated & dynamic environments
Will it make the correct decision when encountering expected, unexpected or unknown situations?
How trustworthy is the information, given its current situational awareness?
How to prevent undesired emergent behavior, as systems interact?
• Provide assurance for machine intelligence and decision-making in complex, uncertain, and dynamic environments
• Develop methods to ensure reliability of human-machine communication and interaction
•Develop rigorous and verifiable architecture systems for data centric autonomous systems
•Develop methodology to V&V fractionated and composable systems
24
Trust & Certification
Trust & Certification: Key Challenges
• Insufficient tools to V&V highly complex, software-intensive systems
• Adaptive/learning systems and uncertain environments yield “near infinite state” systems
• System composition results in potentially hazardous emergent behavior
• Engaging a national team of expertise across DoD, NASA, NSF, DoT, etc. to develop new software certification methods, enabling greater degrees of trust in highly complex, software intensive autonomous systems
• “Design for Certification” asks how: – to supplement test with formal verification – to automate test case generation / reduction
(“Non-Statistical DoE for Software”) • Formal definition and verification of rqmts &
designs to reduce implementation errors and cost in early stages
25
Topics
AFRL Autonomy
Autonomy Challenges in Air Domain
DoD Autonomy Priority Steering Council (PSC) & Community of Interest (COI)
26
Autonomy (Theory or Not)
Human Factors Engineering Machine Algorithmic Development
Partially Observable Markov Decision
Process
Perf
orm
ance
Arousal
Fitts’ Law
?
Strong Weak Low High
Transformed value function for all observations
Yerkes Dodson Law
Theory Linking The Two Sides is Lacking, Mostly Empirical
27
Big Challenges/Issues
• Human-Machine Teaming – Common understanding & perception – Trust in autonomy issues – Shared decision making
• Scalable Teaming – Fractionated and composable systems – Reliable, secure, interoperable communication
• Contested Environments – Situational understanding of the contested environment
– Robust sense and avoid in all conditions
• T&E and V&V – Modeling and simulation – V&V of highly complex, software-intensive systems – Design for certification
28
Questions
Dr. Jim Overholt AFRL Senior Scientist for Autonomy Human Effectiveness Directorate (711HPW/RH) Air Force Research Laboratory [email protected] Office: 937-938-3968 Mobile: 937-829-1179
Kris Kearns AFRL Autonomy Portfolio Lead Human Effectiveness Directorate (711HPW/RH) Air Force Research Laboratory [email protected] Office: 937-656-9758 Mobile: 937-430-4897