ABSTRACT. Losses stemming from energy theft and system faults are a major challenge to providing reliable electrical service in developing areas of the world. Managing these losses is a vital part of ensuring system stability and maintaining a functioning microgrid. Despite this, even losses that are detected are often not addressed by microgrid operators because of the significant hurdle of locating unauthorized loads. In this paper, we propose a method for the localization of unauthorized loads based on the fine-grained sensing of voltage phase across a microgrid. Unlike other approaches, the proposed method utilizes only smart meter sensing and does not rely on expensive inline power metering. Voltage phase measurements feed a graphical model of a power distribution network, which yields the locations of loads as they connect to the system. We evaluate our method using a circuit-based approach in SPICE by simulating loads on a real-world microgrid topology. We then validate our simulation results on a laboratory microgrid testbed using real loads, showing that fine-grained voltage sensing can be effectively leveraged to localize unauthorized loads in microgrids.

ABSTRACT. Recent attacks against cyber-physical systems (CPSes) show that traditional reliance on isolation for security is insufficient. This paper develops real-time assessment and mitigation of an attack’s impact as a system’s built-in mechanisms. We focus on a general class of attacks, which we call time delay attack, that delays the transmissions of control data packets in a linear CPS control system. Our attack impact assessment, which is based on a joint stability-safety criterion, consists of (i) a machine learning (ML) based safety classification, and (ii) a tandem stability-safety classification that exploits a basic relationship between stability and safety, namely that an unstable system must be unsafe whereas a stable system may not be safe. The ML addresses a state explosion problem in the safety classification, whereas the tandem structure reduces false negatives in detecting unsafety arising from imperfect ML. We apply our approach to assess the impact of the attack on power grid automatic generation control, and accordingly develop a two-tiered mitigation that tunes the control gain automatically to restore safety where necessary and shed load only if the tuning is insufficient. Extensive simulations based on a 37-bus system model are conducted to evaluate the effectiveness of our assessment and mitigation approaches.

ABSTRACT. This paper proposes a first-principle model of GPS receivers, that allows us to exploit the trade-off between battery consumption and positioning accuracy. We present the model and propose a GPS sampling strategy that uses both the current positioning confidence, and information about the GPS status. We complement the GPS sensor with internal measurement units and show how the given model exposes the battery-accuracy trade-off in the context of sensor fusion. We demonstrate the usefulness of the proposed sampling strategy using both simulation and real data.

ABSTRACT. Reliable operation of mobile devices, such as smartphones and tablets, has become essential for great many users around the globe. Mobile devices, however, have been reported to su er from frequent, unexpected shutoffs --- e.g., shu ing o even when their batteries were shown to have up to 60% remaining state-of-charge (SoC) --- especially in cold environments. Their main cause is found to be the inability of commodity mobile devices to account for the strong dependency between battery SoC and the environment temperature. To remedy this problem, we design, implement, and evaluate EA-SoC, a real-time Environment-Aware battery SoC estimation service for mobile devices. EA-SoC estimates the battery SoC by predicting the end-of-discharge battery resistance, grounding on (1) a thermal circuit model that describes the interactions among the battery’s discharge current, temperature, and the environment, and (2) an empirically validated data-driven model on the relations between battery temperature and resistance. We have conducted 35 experimental case-studies with two Nexus 5X smartphones to evaluate EA-SoC. EA-SoC is shown to report an average of 3% SoC when the phones shut o even in a −15o C environment, while that reported by the phones’ built-in fuel-gauge chips could be over 90%.

ABSTRACT. In this paper, we propose a model-free reinforcement learning method to synthesize control policies for mobile robots modeled as Markov Decision Process (MDP) with unknown transition prob- abilities that satisfy Linear Temporal Logic (LTL) specifications. Specifically, we develop a reduced variance deep Q-Learning tech- nique that relies on Neural Networks (NN) to approximate the state-action values of the MDP and employs a reward function that depends on the accepting condition of the Deterministic Ra- bin Automaton (DRA) that captures the LTL specification. The key idea is to convert the deep Q-Learning problem into a nonconvex max-min optimization problem with a finite-sum structure, and develop an Arrow-Hurwicz-Uzawa type stochastic reduced vari- ance algorithm with constant stepsize to solve it. Unlike Stochastic Gradient Descent (SGD) methods that are often used in deep rein- forcement learning, our method can estimate the true gradients of an unknown loss function more accurately, improving the stability of the training process. Moreover, our method does not require learning the transition probabilities in the MDP, constructing a product MDP, or computing Accepting Maximal End Components (AMECs). This significantly reduces the computational cost and also renders our method applicable to planning problems where AMECs do not exist. In this case, the resulting control policies minimize the frequency with which the system enters bad states in the DRA that violate the task specifications. To the best of our knowledge, this is the first model-free deep reinforcement learning algorithm that can synthesize policies that maximize the probability of satis- fying an LTL specification even if AMECs do not exist. Rigorous convergence analysis and rate of convergence is provided for the proposed algorithm as well as numerical experiments that validate our method.

ABSTRACT. Machine learning (ML) techniques are increasingly applied to decision-making and control problems in Cyber-Physical Systems (CPS) among which many are safety-critical, e.g., chemical plants, robotics, autonomous vehicle. Despite the significant benefits brought by ML techniques, they also raise additional safety issues because 1) most expressive and powerful ML models are not transparent and behave as a black box and 2) the training data which plays a crucial role in ML safety is usually incomplete. An important technique to achieve safety for ML models is “Safe Fail”, i.e., a model selects a reject option and applies the backup solution, a traditional controller or a human operator for example, when it has very low confidence in a prediction.

Data-driven models produced by ML algorithms learn from train- ing data, and hence they are only as good as the examples they have learned. In other words, the training data constrains the ML models’ performance. As observed in many previous studies, a feature space that lacks training data generally has a much higher error rate than the one that contains sufficient training samples [29]. Therefore, it is essential to identify such an error-prone feature space and reject the unreliable predictions made there. In this paper, we propose an efficient Feature Space Partitioning Tree (FSPT ) to address this problem. Using experiments, we also show that, in general, ML models have a higher loss/error rate in feature space partitions with lower FSPT scores. By rejecting predictions with very low FSPT scores, the prediction loss/error rate can be reduced significantly.

ABSTRACT. Identification of patterns from discrete data time-series for statistical inference, threat detection, social opinion dynamics, brain activity prediction has received recent momentum. In addition to the huge data size, the associated challenges are, for example, (i) missing data to construct a closed time-varying complex network, and (ii) contribution of unknown sources which are not probed. Towards this end, the current work focuses on statistical neuron system model with multi-covariates and unknown inputs. Previous research of neuron activity analysis is mainly limited with effects from the spiking history of target neuron and the interaction with other neurons in the system while ignoring the influence of unknown stimuli. We propose to use unknown unknowns, which describes the effect of unknown stimuli, undetected neuron activities and all other hidden sources of error. The generalized linear model links neuron spiking behavior with past activities in the ensemble neuron system, as well as the unknown influence. The maximum likelihood estimation with the fixed-point iteration method is implemented. The fixed-point iterations converge fast, and besides, the proposed methods can be efficiently parallelized and offer computational advantage especially when the input spiking trains are over long time-horizon. The developed framework provides an intuition into the meaning of having extra degrees-of-freedom in the data to support the need for unknowns. The proposed algorithm is applied to simulated spike trains and on real-world experimental data of mouse somatosensory, mouse retina and cat retina. The model shows a successful increasing of system likelihood with respect to the conditional intensity function, and it also reveals the convergence with iterations. Results suggest that the neural connection model with unknown unknowns can efficiently estimate the statistical properties of the process by increasing the network likelihood.

ABSTRACT. In this paper, we aim to optimize the process of Connected and Automated Vehicles (CAVs) merging at a traffic intersection while guaranteeing the state, control, safety and safe merging constraints. We decompose the task of automatic merging for all the CAVs in a control zone around a merging point into same-lane safety constraints and different-lane safe merging, and implement these requirements using control barrier functions (CBFs). We consider two main objectives. First, to minimize travel time, we make the CAVs reach the road maximum speed with exponentially stabilizing control Lyapunov functions (CLF). Second, we penalize energy consumption as a cost in an optimization problem. We then decompose the merging problem into decentralized subproblems formulated as a sequence of quadratic programs (QP), which are solved in real time. Our simulations and comparisons show that the method proposed here outperforms ad hoc controllers used in traffic system simulators and provides comparable results to the optimal control solution of the merging problem in earlier work.

ABSTRACT. In this paper we describe an experience report and field deployment of real-time filtering algorithms used with a robotic vehicle to smooth emergent traffic waves. When smoothing these waves in simulation, a common approach is to implement controllers that utilize headway, relative velocity and even acceleration from smooth ground truth information, rather than from realistic data. As a result, many results may be limited in their impact when considering the dynamics of the vehicle under control and the discretized nature of the laser data as well as its periodic arrival. Our approach discusses trade-offs in estimation accuracy to provide both distance and velocity estimates, with ground-truth hardware-in-the-loop tests with a robotic car. The contribution of the work enabled an experiment with 21 vehicles, including the robotic car closing the loop at up to 8.0 m/s with this filtered estimate, stressing the importance of an algorithm that can deliver real-time results with acceptable accuracy for the safety of the drivers in the experiment.

ABSTRACT. Using deep reinforcement learning, we train control policies for autonomous vehicles leading a platoon of vehicles onto a roundabout. Using Flow, a library for deep reinforcement learning in micro-simulators, we train two policies, one policy with noise injected into the state and action space and one without any injected noise. In simulation, the autonomous vehicle learns an emergent metering behavior for both policies in which it slows to allow for smoother merging. We then directly transfer this policy without any tuning to the University of Delaware Scaled Smart City (UDSSC), a 1:25 scale testbed for connected and automated vehicles. We characterize the performance of both policies on the scaled city. We show that the noise-free policy winds up crashing and only occasionally metering. However, the noise-injected policy consistently performs the metering behavior and remains collision-free, suggesting that the noise helps with the zero-shot policy transfer. Additionally, the transferred, noise-injected policy leads to a 5% reduction of average travel time and a reduction of 22% in maximum travel time in the UDSSC. Videos of the controllers can be found at https://sites.google.com/view/iccps-policy-transfer.