1. HERACLEIA Human Centered Computing Lab,
A Smart Assessment & Personalized Training Service to Enhance Performance and Safety in the Factory of the Future, Industry 4.0
Kostas Tsiakas, Michalis Papakostas, Maher Abujelala and Fillia Makedon
HERACLEIA Human Centered Computing Lab,
CSE Dept. UTA

2. The Problem, Need and Industrial Relevance
Factory of the Future and Industry 4.0
Changing landscape in manufacturing: Shift from automated to intelligent manufacturing with the use of Robots
Need to train or retrain human workers to work seamlessly and safely with robots in manufacturing settings (industry 4.0)
Human Participation is integral to designing effective HRI systems
Socially Assistive Robots (SAR) can provide personalized training In designing intelligent manufacturing systems

3. Two critical questions for Factory of the Future (FoF)
”How can we enable factory workers to modify the way in which robots react and collaborate with them in the workplace?”
and
”How can we enable robots to sense and adapt to the physiological and cognitive conditions of their human co-worker to ensure safety and high performance?”

4. Project Goals
Develop a Smart Assessment and Training Service called iWork
Monitor and assess worker skills while performing a simulated manufacturing/ training task
Collect and analyze multisensing data while robot provides feedback: errors made, time to complete task, cognitive and physical obstacles, safety risks.
Research Objective: Interactive Personalization using Interactive Reinforcement Learning

5. Research Methods and Novelty of Approach
The HUMAN CENTRIC CPS system includes 3 components (human, physical and computational). Human user can have different roles (worker, supervisor). Task-dependent multisensing HRI data are collected while worker interacts and collaborates with the robot. These data are analyzed and used by the computational component (machine learning, control algorithms, reinforcement learning) to make the appropriate adjustments towards adaptation. Intelligent monitoring and control interfaces can be used to integrate human input to the process (e.g., a human controller who supervises the interaction).

6. Service Architecture Procedure Steps Worker Assessment Data Analysis
Recommendations
Human-in-the-Loop
Worker Training
The HUMAN CENTRIC CPS system includes 3 components (human, physical and computational). Human user can have different roles (worker, supervisor). Task-dependent multisensing HRI data are collected while worker interacts and collaborates with the robot. These data are analyzed and used by the computational component (machine learning, control algorithms, reinforcement learning) to make the appropriate adjustments towards adaptation. Intelligent monitoring and control interfaces can be used to integrate human input to the process (e.g., a human controller who supervises the interaction).

7. Outcomes and Deliverables
Provide a set of Experimental Testbeds for different factory jobs:
simulate manufacturing tasks in order to assess and train for cognitive, physical and collaborative human-robot skills
Provide Multisensing Data Analysis for non technical users to monitor Task performance, engagement, fatigue.
Apply novel Interactive Reinforcement Learning methods that guide personalization by including implicit worker engagement/attention data (EEG) and factory supervisor expertise.

8. Developed Experimental Testbeds

9. Developed Experimental Testbeds

10. Use Case: Data Collection and Analysis
Collect data during a cognitive training task (sequence learning)
Identify human behavior patterns in terms of performance, skills and engagement
Utilize these observed behaviors to provide personalized training – INTERACTIVE RL

11. Next Steps and Ongoing Work
Evaluate machine learning methods for multisensing behavior monitoring
Focus on specific aspects, i.e., attention, cognitive workload, engagement
Develop a battery of assessment and training tasks
Context-specific
Data Collection and Analysis
Examine relations between task performance and cognitive and/or physical aspect
Interactive Reinforcement Learning through multisensing
User Studies

12. Project Timeframe and Budget
2 year project duration
Year 1: Hardware and software setup and preliminary tests
Year 2: Task demonstration and evaluation with real subjects
Estimated cost per year: $55K
$15K materials and equipment;
$40 K labor (PI, Graduate student assistant)
TOTAL: $55K x 2= $110K

13. Impact and Vision of proposed work
Transform vocational training field
Create a highly marketable product applicable to numerous industries
Enable targeted hiring with low-cost training
Introduce robots to manufacturing training in ways that increase productivity, and safety
Help thousands of persons seeking a job in the factory of the future

14. Related Publications
K.Tsiakas, M. Papakostas, M. Theofanidis, S. Wang, M. Burzo, R. Mihalcea, M. Bell, F. Makedon, “An Interactive Multisensing Framework for Personalized Human Robot Collaboration and Assistive Training Using Reinforcement Learning”, PETRA 2017
Christopher Collander, Joseph Tompkins, Alexandros Lioulemes, Michail Theofanidis, Ali Sharifara, Fillia Makedon, An Interactive Robot-based Vocational Assessment Game using Lego Assembly, The 1st Workshop on Assistive Robots , PETRA 2017.
K. Tsiakas, C. Abellanoza, M, Abujelala, M. Papakostas, T. Makada, F. Makedon, “Towards Developing a Socially Assistive Robot for Personalized Cognitive Training”, in Workshop ‘17
K. Tsiakas, M. Abujelala, A. Lioulemes, F. Makedon, "An Intelligent Interactive Learning and Adaptation Framework for Robot-Based Vocational Training", in IEEE Symposium Series in Computational Intelligence, C2RAT, SSCI 2016
K. Tsiakas, M. Dagioglou, V. Karkaletsis, F. Makedon., "Adaptive Robot Assisted Therapy using Interactive Reinforcement Learning", in International Conference on Social Robotics, ICSR 2016

15. WWW.PETRAE.ORG about corfu: https://atcorfu.com/corfu-island/

16. Thank you makedon@uta.edu