Autonomy will become central to future Cyber-Physical Systems, as we scale existing systems and connect multiple systems together to address challenges in domains such as health, transport and energy. Modern artificial intelligence and machine learning technologies promise to improve every aspect of CPS such as maintenance, planning, optimization. CPSes can continually improve based on experience and adapt to changing circumstances.

Machine learning is already being used for various aspects of CPS, from recognizing human activities to robotic control. Much of the prior work has focused on supervised learning, where labels for each input are provided by a human, and a model is trained that learns the patterns in the data. However, in many CPS scenarios, these labels are expensive to collect on a large scale (e.g., autonomous driving) or not known beforehand (e.g., power transmission levels). This workshop seeks contributions where the model is continuously learning while the system is deployed using methods such as model predictive control, reinforcement learning, and behavioral cloning. The CPS community is slowly adopting these methods with applications to autonomous driving, HVAC control, medical intervention, and urban planning. Autonomous CPSes bring in a variety of research challenges: specification of constraints such as traffic rules, safety, high-assurance, risk-sensitive behavior, explainability of the model, transfer from simulation of the physical world to the real-world, multi-agent coordination. While these topics are being studied in robotics conferences such as CoRL, they are only beginning to be explored in the much richer CPS domain and lack a current focused publication forum.

Call for Paper (flyer)

The Workshop on Autonomy in Cyber-Physical Systems at the CPS-IoT Week 2020 seeks to bring together researchers to create solutions for the development of autonomous cyber-physical systems that can continually learn while being deployed in real physical and human environments. As CPS applications are typically safety-critical, we seek contributions that address safety and reliability concerns. As a goal of the workshop is to build a community of CPS researchers who are interested in frameworks, algorithms, tools, platforms, and testbeds for the development of autonomous cyber-physical systems, we seek contributions across disciplines - continual learning, reinforcement learning, control, human-machine interaction, safety, reliability, and verification. The workshop in particular encouraged submissions that propose and explore new ideas as opposed to incremental research on established ideas.

Topics of interest

• Continual learning in physical spaces
  • Reinforcement learning
  • Imitation learning / behavioural cloning
  • Model predictive control
• Safe and efficient exploration
• Transfer of models from simulation to real-world
• Application spaces beyond robotics
• Issues in partially autonomous CPS
• Objective specification
• Learning in the presence of constraints
• Multi-agent and multi-objective systems
• Interpretability / explainability of policy
• Robustness to system changes
• Verification of automation policies
• Tools and platforms for autonomous learning

Important dates

• Paper due: Jan 17 AoE (extended)
• Decision notification: February 5
• Camera Ready: Feb 14 AoE

Invited speakers

Stay tuned. Will be announced soon.

Organizers

Workshop Chairs

• Yasser Shoukry, University of California – Irvine (UCI)
• Bharathan Balaji, Amazon
• Mani Srivastava, University of California – Los Angeles (UCLA)

Website Chair

• Achin Jain, University of Pennsylvania

Publicity Chair

• Chuchu Fan, University of Illinois Urbana-Champaign / Massachusetts Institute of Technology

Technical Program Committee

• Alberto Cerpa, University of California – Merced
• Alberto Elfes, CSIRO
• Archan Misra, Singapore Management University (SMU)
• Enes Bilgin, Microsoft
• Jorge Ortiz, Rutgers University
• Luis Garcia, University of California – Los Angeles (UCLA)
• Mehmet Köseoğlu, Hacettepe University
• Miroslav Pajic, Duke University
• Murali Balakrishnan Narayanaswamy, Amazon
• Olaf Landsiedel, Kiel University
• Paulo Tabuada, University of California – Los Angeles (UCLA)
• Peyman Moghadam, CSIRO
• Pratik Chaudhari, University of Pennsylvania
• Rahul Mangharam, University of Pennsylvania
• Raghupathy Sivakumar, Georgia Institute of Technology
• Salma Elmalaki, University of California – Irvine (UCI)
• Saman Halgamuge, University of Melbourne
• Susmit Jha, SRI International
• Takashi Okoshi, Keio University
• Tarek Abdelzaher, University of Illinoise Urbana-Champaign
• Valerie Liptak, Amazon
• Wen Hu, Universty of New South Wales
• Yonggang Wen, Nanyang Technological University

Steering Committee

• John Stankovic, University of Virginia
• Rajesh Gupta, University of California – San Diego (UCSD)
• Sanjit Seshia, University of California – Berkeley (UCB)

Sponsors

Amazon

Contact Information

The workshop organizers can be reached at chairs2020@autocps.org.