Cognitive Autonomy for Human CPS

[2019- ongoing]

This Project seeks constructive methods to answer the question: How can we design cyber-physical systems to be responsive and personalized, yet also provide high-confidence assurances of reliability? Cyber-physical systems that adapt to the human, and account for the human’s ongoing adaptation to the system, could have enormous impact in everyday life as well as in specialized domains (biomedical devices and systems, transportation systems, manufacturing, military applications), by significantly reducing training time, increasing the breadth of the human’s experiences with the system prior to operation in a safety-critical environment, improving safety, and improving both human and system performance.

This research investigates a human-centric architecture for “cognitive autonomy” that couples human psycho-physiological and behavioral measures with objective measures of performance. The architecture has four elements:

1) A computable cognitive model which is amenable to control, yet highly customizable, responsive to the human, and context dependent

2) A predictive monitor, which provides a prior probabilistic verification as well as real-time short-term predictions to anticipate problematic behaviors and trigger the appropriate action

3) Cognitive control, which collaboratively assures both desired safety properties and human performance metrics

4) Transparent communication, which helps maintain trust and situational awareness through explanatory reasoning.

For the first stage of this project, we have been motivated to develop a new human-automation interaction scheme, which is focused on adaptation and customization for different human users in terms of their skill level. Consider a driving assistance scenario as an example to illustrate this with different drivers with different driving skills; for a skilled driver, it is expected to give more authority to the driver because he/she may need little assistance from the automation, meanwhile it is desired to enable the automation to learn from the skilled driver. For a novice driver, the automation is supposed to take a more active role and it is expected for the automation to provide the assistance that can help the novice to emulate the performance of skilled drivers.

Figure 1: Proposed scheme for human-automation interaction

A scheme for human-automation interaction has been proposed and demonstrated by illustrative experiment (see Fig. 1 for a scheme and Fig. 2 for an experiment platform; the experiment results will be posted after December 2020). The automation interacts with human users in two ways: learning from the human expert in terms of mission objective awareness and assisting the human novice to emulate the mission objective of the expert. Especially, the novelty of this work comes from the adaptation and customization of the control parameter for each individual novice. This is achieved by obtaining the quantified mission objective of the expert and that of the novice using the inverse optimal control or inverse reinforcement approaches. The automation configures the controller that allows the novice to emulate the objective of the expert using the Pontryagin differentiable programming (PDP) based gradient approach.

Fig. 2: Human Subject Experiment Tool (Penguin Game)

We are also planning for further studies not only developing theoretical approaches but also human subject experiments with cognitive state measurements, as a part of collaborative works between multi-institute research. The future works would be human intent inference, iterative trajectory optimization, and safe reach control for the human-automation interaction framework.

Principal Investigators list

Related Publications

S. Byeon, W. Jin, D. Sun, and I. Hwang, “Multi-Mode Human-Automation Interaction for Novice Assistance,” IEEE Conference on Decision and Control 2020