Cyber Physical System Challenges for Human-in-the-Loop Control

Sirajum Munir, John A. Stankovic, Chieh-Jan Mike Liang, Shan LinFeedback Computing 2013 · OpenReview

TL;DR

This paper identifies three critical challenges in building effective human-in-the-loop (HITL) cyber-physical systems: a lack of comprehensive understanding of HITL control types, the difficulty in modeling human behavior, and the challenge of formally integrating these human models into control system design. Addressing these is crucial for advancing HITL systems in areas like energy management, healthcare, and autonomous driving.

What Problem Does This Paper Address?

Cyber-physical systems (CPS) are increasingly integrating humans into their control loops to leverage human intuition, adaptability, and decision-making. This "human-in-the-loop" (HITL) approach offers significant potential for applications ranging from energy management and healthcare to automotive systems. However, effectively incorporating humans into control systems is far from straightforward. The core problem this paper addresses is the lack of a structured framework and established methodologies for designing and implementing robust HITL CPS. Specifically, it highlights the need to:

  1. Understand the diverse landscape of HITL controls: HITL systems aren't monolithic. They range from humans directly commanding a system to systems passively monitoring humans and acting upon their behavior. A clear taxonomy is needed to categorize these variations and identify common principles.
  2. Model human behavior: Humans are complex, exhibiting physiological, psychological, and behavioral nuances that are difficult to capture mathematically. Existing system identification techniques are often insufficient for this task, requiring extensions or entirely new approaches.
  3. Formally integrate human models into control theory: Even with a human behavior model, determining where and how to integrate it into a formal feedback control framework is a significant hurdle. This includes guaranteeing system stability, accuracy, and performance as human behavior evolves.

Methods

The paper doesn't propose a single novel algorithm or technique but rather articulates a research agenda by identifying and elaborating on three fundamental challenges. The "methods" discussed are conceptual and focus on the need for new approaches:

  • Taxonomy Development: The authors propose creating a taxonomy of HITL applications based on the type of control employed:
  • Humans directly control the system (e.g., supervisory control where humans intervene or provide commands).
  • Systems passively monitor humans and take appropriate actions (e.g., health monitoring, smart home systems).
  • Hybrid systems combining both direct control and passive monitoring.
  • Extensions to System Identification: The paper calls for extending existing system identification techniques or developing new ones to derive models of human behavior. This includes exploring how to define inputs, states, and state transitions for human models, considering physiological, psychological, and environmental factors.
  • Formal Integration Frameworks: The authors emphasize the need to develop formal methodologies for incorporating human behavior models into feedback control. This involves exploring where human models can be placed within a control loop (e.g., inside the controller, as part of the system model) and developing theories to guarantee system properties like stability and accuracy. Techniques like clustering, data mining, inference, and first-principle models are suggested as potential avenues.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • HITL is not a one-size-fits-all solution: The paper stresses that understanding the spectrum of HITL controls is crucial. Different applications require different levels of human involvement and different modeling approaches.
  • Human behavior is a moving target: Modeling human behavior is inherently difficult because it's dynamic, influenced by a multitude of factors, and can change over time. This necessitates adaptive control strategies.
  • Formal guarantees are essential for safety-critical HITL systems: For applications like autonomous driving or healthcare, simply having a human "in the loop" isn't enough. We need formal methods to ensure the system's stability and predictable behavior, even as human behavior evolves.
  • Bridging the gap between behavioral science and control theory is key: The paper highlights the need to integrate insights from psychology, physiology, and behavioral economics with rigorous control engineering principles.

Key Results

This paper is a position paper outlining research challenges, not an empirical study with quantitative results. Its primary "results" are the identification and articulation of the three core challenges:

  1. Challenge 1: The need for a comprehensive understanding and taxonomy of HITL control types.
  2. Challenge 2: The necessity for extensions to system identification or new techniques to derive models of human behaviors.
  3. Challenge 3: The critical requirement to incorporate human behavior models into the formal methodology of feedback control, ensuring formal performance guarantees.

Limitations and Future Work

The authors explicitly acknowledge the limitations and suggest future work by framing these challenges as open research problems:

  • Limited understanding of HITL spectrum: The paper calls for further work in creating and refining taxonomies for HITL applications.
  • Difficulty in human behavior modeling: Future work needs to focus on developing robust and adaptable modeling techniques for human behavior, potentially combining various data mining, inference, and first-principle approaches.
  • Formal integration challenges: A significant area for future research is developing new theoretical frameworks and control strategies that can formally guarantee the stability and performance of HITL systems as human behavior evolves. This includes exploring how to define and learn these properties. The paper suggests that adaptive control with humans-in-the-loop will be necessary.

Summary updated

BibTeX
@inproceedings{munir2013cyber,
  author = {Sirajum Munir and John A. Stankovic and Chieh-Jan Mike Liang and Shan Lin},
  title = {{Cyber Physical System Challenges for Human-in-the-Loop Control}},
  booktitle = {Feedback Computing},
  year = {2013},
  url = {https://openreview.net/forum?id=tmElaXz0XL},
}

Loading related content...

Figure