The International Conference on Cyber-Physical Systems 2020


On January 1st, an artificial intelligence system was capable of detecting breast cancer better than human experts. February and March both witnessed discoveries about life in the universe: in fact the organic molecules detected by the Curiosity rover are consistent with the hypothesis of early life on Mars and primitive microorganisms may have existed on the planet Mercury. These are just some examples of how computation is changing the world we live in and our knowledge. Ultimately, these are examples of how the cyber and physical world are connected.

But 2020 is a special year, also because of the 2019-20 coronavirus pandemic. In the midst of the corona crisis, computation can help us understand what is the best course of action, given a model and a set of hypotheses of how the virus spreads. For example, this video by Grant Sanderson (3blue1brown) shows how to simulate (with some modeling assumptions) an epidemic and what is the effect of different measures and of when they are taken.

In the middle of these exciting and scary times, we prepared the program for the 11th IEEE/ACM International Conference on Cyber-Physical Systems, collecting scientific contributions on the intersection between the cyber and physical aspects of our world. The main conference program is composed of 7 sessions, touching upon various topics including security, learning, and medicine.

Session 1

We start with a session on how to guarantee the security of cyber-physical systems. The first paper looks at the privacy of dataset that are evolving over time. The repeated release of information at successive timesteps leads to the loss of differential privacy. The author relaxes the standard definition of differential privacy to a newly proposed discounted differential privacy. The second paper provides an analytical model for dynamic information flow tracking. The model is used to (optimally) select locations in the system to perform security analysis. Optimally here means that the selected locations maximize the probability of detection of the attack, while minimizing the cost to detect it. The third paper proposes to use a signaling game with evidence, in a digital-twin architecture, to secure a class of control systems against data-integrity attacks on estimated state. The fourth paper presents an approach to detect cyber-attacks in cyber-physical systems, which relies on a model of the physics of the system and the code supplied by the system operator.

Session 2

The second session is about smart infrastructures, like factories and automated traffic intersections. The first paper formalizes a subset of traffic rules  from the California driving manual using first-order logic. These rules are about yielding in uncontrolled intersections, with or without signs or traffic lights. The second paper introduces a new spatio-temporal logic to reason about data coming from smart city sensors. In this new logic, the introduction of aggregation and counting spatial operators helps the authors to formalize a large corpus of smart city requirements. The third paper presents an algorithm to allocate charging slots to electric vehicles. The slots are allocated based on the reputation of the electric vehicle. The reputation is in turn decided based on the past behavior of the vehicles in reporting their behavior truthfully.

Session 3

Here, we cover a topic that became more and more important in recent years: learning. The first paper deals with transfer reinforcement learning, which means transferring knowledge from a demonstrator agent that has learned something about a task to a learner. The demonstrator collects data and uses it to calculate the upper and lower bounds for value functions. The learner can then project predictions onto these bounds allowing for faster learning. The second paper presents a method to infer environment assumptions from labelled data about relevant variables, in the form of a signal temporal logic formula. Given input and output traces, the authors mine an input specification so that the negative examples of the dataset do not satisfy this input specification, and the positive examples of the dataset satisfy it. The third paper develops a method providing optimality guarantees for controller synthesis in continuous-space Markov decision processes, with unknown dynamics. The approach first computes a finite approximation to the original model, and then uses the classical convergence results for reinforcement learning techniques to determine the controller that maximizes the probability of satisfaction over the unknown real mode, while providing probabilistic closeness guarantees.

Session 4

This session is about both medicine and the internet of things. The first paper presents a study building a reinforcement learning model for synthesis optimal temporal deep brain stimulation controller for Parkinson’s disease. The second paper presents a technique for non-invasive measurement of fetal heart rates. The main challenge in this problem is to separate the fetal signal from the maternal signal. The third paper presents a dynamic network resource allocation protocol that matches reserved network bandwidth to the communication requirements for tactile applications, given current physical motion speed, meaning the speed of the hand motion. The fourth paper proposes a new method to obtain and update Wi-Fi fingerprint maps for indoor localization. The key idea is to use an autonomous robot that collects Wi-Fi signals in an indoor space without stopping to reduce the time and energy cost for maintaining accurate fingerprint maps compared with existing methods that need to stop for several seconds at many locations.

Session 5

The next session is dedicated to outstanding papers that received the best reviewers’ evaluation. The first of these papers explores questions around the software implementations of controllers. The authors mine the commit data from two open source controller software repositories, which do not use model-based design. The paper then explore the degree to which controllers’ software evolves over time, and the types of changes that could not be reflected in typical control design models. Finally, they modify a code mutation engine, to introduce similar changes into code that is automatically generated for three examples of Simulink-based controllers and investigate the extent to which such software mutations can cause the output of the implemented controller to change significantly. The second outstanding paper considers the problem of synthesizing a controller for a stochastic system subject to a metric interval temporal logic objective, which extends linear temporal logic to include timing constraints. In the synthesis problem, an adversary is able to affect actuator inputs and the time-stamp that is read by the defender. The synthesis problem is transformed into a problem of reaching accepting components in the product game, and the paper shows how this problem can be solved via a value-iteration procedure. The third outstanding paper presents a novel approach for online detection of out of distribution samples in learning-enabled CPS. It applies conformal anomaly detection on high dimensional data for this purpose. This allows the authors to detect anomalies in the execution of cyber-physical systems, by calculating the non-conformity score, or the amount of difference between the given sample that is obtained from the real execution and priorly acquired training samples.

Session 7

This session is about estimation and control. The first paper in the session proposes an intermittent connectivity control framework for robot teams that live in communication constrained environments. The intermittent connectivity problem is formulated as an integer linear program that also allows for synthesizing plans for auxiliary goals, like the exploration of an area. The second paper presents a distributed estimation algorithm to be used in wireless networks with constrained resources. The authors argue that classical solutions such as the distributed Kalman filter require too much computation and communication. They then proceed to propose an event-version diffusion Kalman filter in which communication only occurs when the uncertainty (measured through each local covariance matrix) gets too high. The third paper focuses on developing a robot motion planner which generates collision-free, dynamically feasible, and socially aware trajectories for robots operating in close physical proximity with humans. The definition of a socially aware trajectory is a trajectory which minimizes the total number of human complaints received while the trajectory is actuated. The fourth paper in the session presents an approach to reengineer existing closed loop discrete time linear time-invariant systems to form an optimization problem. The result of the optimization problem improves the performance of the controller by just adding a control term to the original loop, meaning that the original controller can be upgraded easily.

Session 8

The last session is dedicated to tools and co-design methods. The first paper presents a protocol for building a replicated controller on a distributed network. Replication improves the original controller with two main features, state- and output-consistency. This means that the outputs applied to the actuators are always based on an internal state that is reachable from all other previous internal states that also previously produced an output to the actuators, and that in each control cycle it cannot happen that two replicas will send different output values to the replicas. The second paper presents an approach that makes remote attestation feasible in resource constrained internet-of-things devices without requiring hardware modifications while providing security guarantees that are better than traditional software-based approaches. The approach uses software-based memory isolation and formal verification to guarantee that devices that have been compromised can be identified. The third paper proposes an approach which automatically generates semantically correct real-time implementations of hybrid systems that can be analyzed for their worst-case timing for safety-critical domains. The authors propose a methodology of implementing hybrid systems in hardware, guaranteeing strict timing constraints.

Is that all?

You might have noticed that there is no Session 6 in this list. That’s because session 6 contains the work-in-progress papers. Don’t miss a chance to check out what exciting work is ongoing and get inspired!

Due to the current situation, the presentations for all the papers will be video-taped and videos will be available following the links that you have encountered throughout this post. Let me say that I am extremely excited to hear all about the ongoing research, and to engage in the online discussions. It was very difficult to select these papers, because we have received many high quality submissions (in fact, just above 100 papers). I would like to thank the ACM SIGBED blog editors, for allowing me to use this space to tell you about our program, all the program committee members and reviewers that contributed to the evaluation, and finally all the authors that devoted their time and effort to these research problems.

I hope you’ll enjoy the virtual ICCPS 2020!

Author bio: Martina Maggio is a professor at Saarland University, Germany (since April 2020), and an associate professor at Lund University, Sweden (since March 2017). She completed her Ph.D. at Politecnico di Milano, on the applications of control-theoretical tools for the design of computing systems. During her Ph.D. she spent one year as a visiting graduate student at the Computer Science and Artificial Intelligence Laboratory at MIT. She moved to Lund University, where she first was a postdoc and then became an assistant professor. In 2019, she spent a sabbatical year at Bosch Corporate Research in Renningen, Germany, working on the verification and validation of control systems in presence of deadline misses and computational faults.

Disclaimer:  Any views or opinions represented in this blog are personal, belong solely to the blog post authors and do not represent those of ACM SIGBED or its parent organization, ACM.

Leave a comment