Attach-3-MaRDU_IDIQ_SOO _v3.pdf

FA8650-23-S-5725

Attachment 3

AManufacturing Research for Digital Utilization (MaRDU)

Statement of Objectives

Period of Performance (PoP): 63 months total (60 months technical, 3 months reporting)

Technical Point of Contact (TPOC): Dr. Sean Donegan (AFRL/RXMD)

Email: sean.donegan@us.af.mil

Phone: (937) 255-1602

1.0 BACKGROUND: Emerging tools across a variety of digital platforms promise to revolutionize the state-of-the-art in manufacturing practice. The intersection of automation through advanced robotics, artificial intelligence, data acquisition & sensorization, industrial visualization, edge-to-cloud computing, and human-machine teaming offers a rich solution set for delivering the future factory. However, the development and integration of these disparate technologies remains a significant challenge, requiring both fundamental and applied research activities.

2.0 OBJECTIVE: The objective of this effort is to perform research and development to advance the state-of-the-art in digital manufacturing of interest to the USAF. The technical scope includes the broad categories of human-machine teaming, machine intelligence, and human-machine trust. This research program will develop, mature, and deliver fundamental knowledge and technical solutions for digital manufacturing challenges of interest to the Air Force in close working collaboration with the Air Force Research Laboratory’s Digital Manufacturing Research

Team.

3.0 TECHNICAL SCOPE:

3.1 Human-Machine Teaming

The future manufacturing floor will incorporate diverse multi-agent teams that include both human experts and machines, where machines include both hardware platforms (robotics, AR/VR, etc.) and software (algorithms, interfaces, etc.). The Air Force anticipates that these teams will cross multiple application domains in defense manufacturing and warfighter readiness, and require significant research and digital infrastructure development for success.

Research in human-machine teaming will include, but is not limited to, the following technical areas.

3.1.1 Digitization of Human Expertise

Human experts offer significant expertise across industrial manufacturing applications. As manufacturing systems increasingly integrate autonomy, it will become critical to effectively capture and deploy this expertise onto robotic and other non-human systems.

3.1.1.1 Learning from Demonstration

In this effort, the contractor shall investigate methods for using digitally captured human experience as a resource for building autonomous “agents”: deployable model-based systems capable of autonomous actions trained via guidance by human experts.

mailto:sean.donegan@us.af

Techniques involved may include, but are not limited to: deep reinforcement learning, task-based learning, iterative learning, and model predictive control.

3.1.1.2 Capturing Human Manufacturing Experience

In order for human expertise to be an effective training tool, it must be digitally captured and stored so that it is useful for downstream analysis and model integration. This effort involves the research and development of techniques that enable digital capture of human experience. Interfaces used to capture human experience include AR/VR, kinaesthetic robotic learning, and motion capture. Data analytic techniques are critical for making such captured data useful. The contractor shall investigate approaches that construct useful data models of digitized experience. Additionally, the contractor shall develop data storage and retrieval approaches for capturing human experience, and design ontological representations of such data where appropriate.

3.1.2 Human-Machine Interaction

Future manufacturing environments will incorporate direct interfacing between humans and machines. In this context, machines include both hardware systems (e.g., robotic arms, manufacturing equipment, AR/VR, etc.) and software systems (e.g., control software interfaces, AR/VR heads-up displays, etc.).

3.1.2.1 Human-Robot Interfacing and Programming

The contractor shall research and develop techniques for human-robot interfacing, building approaches for direct interactions between humans and robots for manufacturing applications. This area of work includes robotic programming and control in collaborative environments, kinesthetic learning, and integration of combined systems of robots, sensors, and computer resources.

3.1.2.2 Augmented and Virtual Reality in Digital Manufacturing

Communication of digital information in a manufacturing environment is vital for effective information flow. The contractor shall investigate approaches for using augmented reality and virtual reality as tools to effectively communicate complex manufacturing information. This work includes, but is not limited to, environment development for

AR/VR and integration of AR/VR with complete manufacturing systems.

3.1.3 Automation, Autonomy, and Robotic Control

As industrial robots become an integral part of future manufacturing environments, the development of agile approaches for robotic control grows in importance. The contractor shall research and develop algorithms for robust robotic automation and autonomy in manufacturing environments, which are often noisy, uncontrolled, and austere. This area includes both development of such techniques in robotic simulation platforms and deployment of the control approaches onto real robotic systems.

3.1.3.1 Robotic Control and Autonomy for Multi-Agent Teams

Diverse, multi-agent teams are a critical component of future manufacturing environments. The contractor shall conduct research and development of approaches for effective autonomy and communication amongst multi-agent teams, which can include multiple humans and robots.

3.1.3.2 Sensor Feedback, Process Control, and Process Automation

Sensorization of manufacturing systems allows for significant information capture during the lifecycle of a given process. Information captured from sensors can be used to effect process control and drive system automation. The contractor shall research and develop novel approaches for sensorization of manufacturing systems and the integration of sensors with process control and automation.

3.2 Machine Intelligence

Technical problems within digital manufacturing often require algorithmic and mathematical solutions. The development of effective algorithms is vital to the maturation of the digital manufacturing enterprise, and includes both theoretical development as well as practical implementation of algorithms in code. The combination of theoretical algorithm development and implementation is collectively referred to here as machine intelligence. Research in machine intelligence includes, but is not limited, the following technical areas.

3.2.1 Algorithms for Machine Intelligence in Manufacturing

Modern approaches in artificial intelligence and machine learning (AI/ML) offer significant benefits for implementation of machine intelligence in digital manufacturing. This includes fundamental development of novel AI/ML algorithms, implementation of these algorithms in code, and their deployment in manufacturing environments.

3.2.1.1 Artificial Intelligence and Machine Learning

The contractor shall conduct research and development into techniques and algorithms for AI/ML applied to digital manufacturing problems. Techniques of interest include, but are not limited to: supervised learning (e.g., classification, regression, discriminant analysis, decision trees, etc.), unsupervised learning (e.g., clustering, anomaly detection, dimensionality reduction, etc.), reinforcement learning (e.g., on/off policy learning, discrete and continuous action spaces, Monte Carlo approaches, etc.), and natural language processing.

3.2.1.3 Deep Learning Architectures

A critical tool for many methods of modern AI/ML is deep networks. The contractor shall research and develop novel architectures for deep learning, including fully-connected networks, convolutional networks, Bayesian networks, graph networks, etc.

3.2.2 Data Analytics

Classical data analytics provides a vital tool for processing digital manufacturing data in combination with techniques from AI/ML. The contractor shall research and develop such techniques for digital manufacturing contexts.

3.2.2.1 Quantitative Image Processing

Images represent a common data medium for materials and manufacturing information.

The contractor shall research and develop techniques for quantitative image processing, including segmentation, alignment, registration, compression, smoothing, and denoising.

3.2.2.2 Applied Statistics and Optimization

Statistical techniques provide robust tools for analysis of a variety of data types, such as tabular and time series data. Additionally, algorithms for optimization from such data sources provides The contractor shall investigate approaches that include, but are not limited to, time series analysis, classical regression, statistical analysis, data mining, Bayesian approaches, and multi-objective optimization.

3.2.2.3 Data Visualization

As manufacturing data grows in complexity, visualization of such data becomes an increasingly important task. The contractor shall research, develop, and implement approaches for data visualization, to include visualization of real-time data from manufacturing platforms.

3.2.3 Cloud Infrastructure and Databases

Implementation of machine intelligence algorithms for manufacturing environments requires significant computational resources. The contractor shall develop cloud-based infrastructure for the development and deployment of digital manufacturing algorithms, to include the development and maintenance of databases, data repositories, and software repositories. The contractor shall investigate approaches for edge devices for use in austere manufacturing environments.

3.3 Human-Machine Trust

As the digital manufacturing enterprise matures, it becomes increasingly important to develop approaches for assessing and quantifying trust. Algorithmic approaches for digital manufacturing may be termed as “black box”, lacking significant explainability on the surface. To understand the risk associated with novel approaches in digital manufacturing, techniques must be developed that effectively determine human-machine trust.

3.3.1 Explainable AI

As a companion to the development of machine intelligence algorithms, approaches that provide explainability are required to properly assess risk and trust. The contractor shall research and develop tools that provide explainability in digital manufacturing systems.

3.3.2 Certification and Validation

Systems that integrate machine intelligence with autonomous manufacturing equipment are sufficiently novel to require new approaches in certification and validation. The contractor shall research and develop techniques for certifying and validating such integrated systems in manufacturing contexts.

3.3.3 Trust within Multi-Agent Teams

Digital manufacturing involves interaction on diverse multi-agent teams, including both humans and robotic platforms. The contractor shall investigate approaches for building trust in such multi-agent teams when applied to manufacturing tasks.

4.0 SPECIAL CONSIDERATIONS

4.1 Fundamental Research

4.1.1 The work of Section 3 of this SOW is “Fundamental Research” in accordance with

National Security Decision Directive 189, National Policy on Scientific Decision Directive 189, National Policy on Scientific, Technical and Engineering Information, in effect on the date of contract award and the Under Secretary of Defense (Acquisition, Technology, and Logistics) memoranda on Fundamental Research, dated May 24, 2010, and on Contracted Fundamental

Research, dated June 26, 2008. As such, the work of those paragraphs of this SOW:

1) Is excepted from the restrictions on disclosure of information imposed by

DFARS 252.204-7000(a)(3) and

2) Is not "covered defense information" as defined by DFARS 252.204-7012

4.1.2 Application of export controls is dictated by the Department of State and Department of

Commerce.

5.0 DELIVERABLES

5.1 CDRLS A001, A002, A003, A004, A005, A006, A007, A008, A009 and A010.

5.2 Software The contractor shall deliver any software developed under this effort, including source code (A008 where applicable).

5.3 Hardware The contractor shall deliver any hardware developed under this effort, including and testing developed under the overall technical scope.

5.4 Test Samples, Prototypes, and Residual Materials: The contractor shall return to the

Government all test samples, prototypes, and residual materials not consumed in testing at the conclusion of the technical effort.

5.5 Data The contractor shall deliver all relevant data including raw machine output along with machine calibration parameters related to any and all experimentation, demonstration or simulations that result from the program (A009 if applicable).

6.0 BASE SUPPORT

Base support shall be provided during the period of performance of this effort, and shall be available at the following location:

Air Force Research Laboratory (AFRL/RX)

Materials and Manufacturing Directorate

Area B, Building 651-655 Complex

Wright-Patterson AFB, OH 45433-7750

7.0 OPERATIONS SECURITY REQUIREMENTS

The contractor shall participate in all activities associated with the disciplines of the organization’s Industrial Security, Information Security, Personnel Security, Operations Security

(OPSEC), and Antiterrorism programs, following appropriate measures in each program as required for this particular contract. Security measures are required to reduce program vulnerability from successful adversary collection, exploitation of critical information, and violations of export control requirements. The prime contractor shall ensure all subcontractors, if applicable, conform to these requirements as required by the prime contractor.

8.0 PROGRAM PROTECTION PLAN (PPP) REQUIREMENTS

Any potential critical program information (CPI) generated as part of this effort will be reviewed to determine the need for a PPP or to be included as part of an existing PPP.