Safe and Robust Control of Autonomous Cyber-Physical Systems

Safety analyses depend on pre-defined assumptions that will often be wrong in practice, as autonomous systems (e.g., drones and self-driving cars) will inevitably encounter incomplete and possibly erroneous knowledge of the environment and other agents. For instance, a human might be less attentive than normal when facing a self-driving car, presuming that it is the responsibility of the self-driving car to maintain a safe distance. This project shows how autonomous systems can safely and timely reason over the uncertainty of the environment they operate in.

Collaborators: Bruno Sinopoli and Aaron F. Bobick (Washington University in St. Louis)

Funding source: National Science Foundation (USA)

Mitigation Strategies for Autonomous Cyber-Physical Systems

While the technology involved in the construction of autonomous cyber-physical systems is advancing rapidly, mitigation strategies for addressing unanticipated situations (imposed due to faults and cyberattacks) they may face in real-world applications is lagging behind. State-of-the-art detection methods pertain to detection of fault and cyberattack and not their isolation. Also, cyber-level reconfiguration is currently limited to software rejuvenation techniques which cannot handle the broad range of contingencies that arise in real-world applications. This underlines the need for systematic methods for rapid detection and isolation of faults and cyberattacks, and the deployment of effective reconfiguration algorithms, which we pursue in this project.

Collaborators: Bruno Sinopoli and Sanjoy Baruah (Washington University in St. Louis), Ilya V. Kolmanovsky (University of Michigan)

Funding source: National Science Foundation (USA)

Quality-of-Experience Management in Video Streaming Applications

Given the complex Internet video delivery ecosystem and presence of diverse bottlenecks, the bitrate adaptation logic in the client-side video player becomes critical to optimize user experience. In the HTTP-based delivery model that predominates today, videos are typically chunked and encoded at different bitrate levels. The goal of an adaptive video player is to choose the bitrate level for future chunks to deliver the highest possible Quality-of-Experience (QoE), e.g., maximizing bitrate while minimizing the likelihood of rebuffering and avoiding too many bitrate switches. In addition to single-player QoE, QoE fairness across multiple heterogeneous video players (i.e., players with different preferences, different devices, different critical levels, etc) is a key factor that should be addressed when multiplayer share a bottleneck. This project will develop a network-assisted algorithm to improve QoE fairness, which requires no modifications on the client side.

Collaborators: Bruno Sinopoli (Washington University in St. Louis), and Vyas Sekar (Carnegie Mellon University)

Funding source: AT&T Inc. (USA)

Constrained Control Systems

The control of systems that are subject to constraints is one of the main challenges faced by the control system community. Introduced at the end of the 70s, the constrained control literature has been dominated by optimization-based methods. These methods usually lead to very good results in terms of performance, and for this reason they have become the golden standard for the control of many constrained systems. However, methods are typically expensive from the computational viewpoint, especially when applied to nonlinear systems and in the presence of uncertainties. The main goal of this project is to develop optimization-free constrained control schemes, to cope with the above-mentioned problems.

Collaborators: Emanuele Garone (Free University of Brussels), Ilya V. Kolmanovsky (University of Michigan), and Daniel Limon (University of Seville)

Funding source: National Fund for Scientific Research (Belgium), National Science Foundation (USA), and European Regional Development Fund (European Union)

Autonomous Drug Delivery in Closed-Loop Anesthesia

Closed-loop anesthesia systems perform automated administration of anesthetic drugs based on feedback from a measured effect. Closed-loop propofol anesthesia manipulates propofol infusion rates based on feedback of the measured clinical effect on depth of hypnosis to induce and maintain a certain level of anesthesia. The main goal of this project is to develop a control scheme which is robust to inter-patient variability, provides less variability in desired clinical effects compared to manual administration of anesthetic drugs, and ensures patient’s safety.

Collaborators: Guy A. Dumont (University of British Columbia), and Emanuele Garone (Free University of Brussels)

Funding source: National Fund for Scientific Research (Belgium)

Control of Autonomous Energy Systems

A Virtual Power Plant (VPP) is a network of power generating units such as wind farms, solar parks, and combined heat and power units, as well as flexible power consumers and storage systems. The interconnected units are dispatched through the central control room of the VPP but nonetheless remain independent in their operation and ownership. The objective of a VPP is to relieve the load on the grid by smartly distributing the power generated by the individual units during periods of peak load. Additionally, the combined power generation and power consumption of the networked units in the VPP is traded on the energy exchange. However, the complexity of VPPs requires complicated optimization, control, and secure communications. In this project, we plan to investigate two fundamental issues in VPPs: 1) energy management, and 2) cybersecurity.

Collaborators: Bruno Sinopoli (Washington University in St. Louis), Josh Taylor (University of Toronto), and Luca Schenato (University of Padova)

Funding source: United States Department of Energy (USA), and Natural Sciences and Engineering Research Council (Canada)