Ioannis Lambadaris

The technology of autonomous vehicles is advancing rapidly, relying heavily on high-precision simulations and high-performance controllers. This study uses actual driving data to model coupled lateral and longitudinal vehicle dynamics with two neural network (NN) structures: a time-distributed multilayer perceptron (TD-MLP) and a long short-term memory (LSTM). Inputs include accelerator pedal position, brake pressure, steering wheel angle, gear number, and road slope, while outputs are acceleration and yaw rate. In addition, a bidirectional LSTM (Bi-LSTM) predicts vehicle control inputs based on expert driver data. A resolution of the path to be tracked is provided to the NN driver, which generates control commands that are subsequently fed into vehicle NN models. The vehicle’s output states are used to compute the quadratic cost function for trajectory tracking. The minimum cost is identified, and the corresponding control inputs are selected as the optimal inputs for each prediction horizon. A comparative analysis was performed between the neural predictive controller (NPC) algorithm and the traditional nonlinear model predictive controller (NMPC), utilizing the interior point optimizer (IPOPT). Incorporating high-fidelity models within NMPC leads to substantial computational overhead. The NPC method maintains both computational efficiency and is adaptive to road slope disturbances, without the need for a classical adaptive control design, making it suitable for real-time control applications.

We consider a multiple-input multiple-output (MIMO) channel, in which the transmit antennas are partitioned into groups, each with a per-group power constraint (PGPC) and a total power constraint (TPC). We considered two cases: (i) right unitary-invariant, including Rayleigh, fading MIMO channels with perfect channel state information at the receiver (CSI-R), and (ii) massive MIMO channels with perfect CSI at the transmitter and the receiver (CSI-TR). For both cases, we show that the rate-optimal input covariance matrix is diagonal, implying reduced design complexity and independent signaling on each antenna. We derive closed-form expressions for the diagonal entries, i.e., the powers allocated to each antenna. For CSI-R channels, we derive a criterion to identify groups with active PGPCs. Majorization theory and Schur-concavity are used to obtain the optimal power allocations. For CSI-TR channels, we use the Karush-Kuhn-Tucker conditions to show that the PGPCs result in a ceiling profile, causing the rate-optimal power allocations to deviate from standard water-filling. Compared to numerical algorithms, our closed-form expressions are significantly more efficient to compute and guarantee globally optimality. Our analytical findings are validated via numerical experiments.

Under a dramatic increase in mobile data traffic, a promising solution for edge computing systems to maintain their local service is the task migration that may be implemented by means of Autonomous mobile agents (AMA). In designing an optimal scheme for task offloading to AMA, we define a system cost as a minimization objective function that comprises two parts. First, an offloading cost which can be interpreted as the cost of using computational resources from the AMA. Second, a penalty cost due to potential task expiration. To minimize the expected (timeaverage) cost over a given time horizon, we formulate a Dynamic programming (DP). However, the DP Equation suffers from the well-known curse of dimensionality, which makes computations intractable, especially for infinite system state space. To reduce the computational burden, we identify three important properties of the optimal policy and show that it suffices to evaluate the DP Equation on a finite subset of the state space only. We then prove that the optimal task offloading decision at a state can be inferred from that at its adjacent states, further reducing the computational load. We present simulations to verify the theoretical results and to provide insights into the considered system.

Under a dramatic increase in mobile data traffic, a promising solution for edge computing systems to maintain their local service is the task migration that may be implemented by means of Autonomous mobile agents (AMA). In designing an optimal scheme for task offloading to AMA, we define a system cost as a minimization objective function that comprises two parts. First, an offloading cost which can be interpreted as the cost of using computational resources from the AMA. Second, a penalty cost due to potential task expiration. To minimize the expected (time-average) cost over a given time horizon, we formulate a Dynamic programming (DP). However, the DP Equation suffers from the well-known curse of dimensionality, which makes computations intractable, especially for infinite system state space. To reduce the computational burden, we identify three important properties of the optimal policy and show that it suffices to evaluate the DP Equation on a finite subset of the state space only. We then prove that the optimal task offloading decision at a state can be inferred from that at its adjacent states, further reducing the computational load. We present simulations to verify the theoretical results and to provide insights into the considered system.

Truck platooning is an innovative logistics approach to lower operational costs, particularly fuel consumption, while addressing contemporary transportation challenges. While recent studies on truck platooning have emphasized platoons’ energy savings, stability, and safety, there has been limited exploration of platoon formation and control. This paper uses optimal control theory to address the dispatching control of trucks with arriving platoons. In particular, trucks arrive at a highway station while platoons arrive alongside it. The station controls the truck holding and dispatching, where trucks are sent out with or without a platoon. Dispatching trucks with an arriving platoon reduces fuel consumption while waiting for a platoon to arrive increases the dwell time (i.e., transportation delay). We assume that an arriving platoon determines the number of trucks (i.e., the batch size) it can accept. Only a single truck can be dispatched if a platoon is absent. Hence, we formulate the dispatching control problem and derive the optimal policy for the discounted costs and the average cost governing the dispatch of trucks alongside platoons. We proved the optimality of threshold policies. Numerical results for the average cost case are presented. They are consistent with the optimal ones.

Low Earth Orbit Satellite Networks (LSNs) are integral to supporting a broad range of modern applications, which are typically modeled as Service Function Chains (SFCs). Each SFC is composed of Virtual Network Functions (VNFs), where each VNF performs a specific task. In this work, we tackle two key challenges in deploying SFCs across an LSN. Firstly, we aim to optimize the long-term system performance by minimizing the average endto-end SFC execution delay, given that each satellite comes with a pre-installed/cached subset of VNFs. To achieve optimal SFC placement, we formulate an offline Dynamic Programming (DP) equation. To overcome the challenges associated with DP, such as its complexity, the need for probability knowledge, and centralized decision-making, we put forth an online Multi-Agent Q-Learning (MAQL) solution. Our MAQL approach addresses convergence issues in the non-stationary LSN environment by enabling satellites to share learning parameters and update their Q-tables based on distinct rules for their selected actions. Secondly, to determine the optimal VNF subsets for satellite caching, we develop a Bayesian Optimization (BO)-based learning mechanism that operates both offline and continuously in the background during runtime. Extensive experiments demonstrate that our MAQL approach achieves near-optimal performance comparable to the DP model and significantly outperforms existing baselines. Moreover, the BO-based approach effectively enhances the request serving rate over time.

The increased number of breast cancer cases worldwide necessitates the development of early breast abnormality detection techniques. Thermography serves as a promising imaging modality that can be used as an adjunctive tool with mammography for early breast abnormality detection. It can be particularly useful for breast abnormality detection in developing or underdeveloped countries that have a limited number of medical professionals and low-power processing units for diagnosis. Appropriate compression of breast thermal images reduces the data storage expenses and computational complexity of the algorithms for breast abnormality detection using thermography. Therefore, we are motivated to use deterministic compressive sensing for the compression of the red-plane extracted from the breast thermograms and detecting breast abnormality in the compressed domain using the compressed red plane. The deterministic compressive sensing (CS) technique employs a given binary block diagonal matrix (DBBD) that acts as a low-pass filter and downsampler and preserves the features needed for abnormality detection. We propose a bilateral asymmetry analysis-based breast abnormality detection technique in the compressed domain. A performance analysis of compressed domain breast abnormality detection technique with red-plane thermograms compressed using CS and non-CS compression techniques at different compression ratios along with an analysis of computational complexities is presented. A comprehensive analysis of the performance of compressed domain breast abnormality detection is also explored when different types of common medical image noises (Gaussian, salt and pepper, and speckle noise) at different noise levels are present.

In the 6G wireless era, the strategical deployment of Virtual Network Functions (VNFs) within a network infrastructure that optimizes resource utilization while fulfilling performance criteria is critical for successfully implementing the Network Function Virtualization (NFV) paradigm across the Edge-to-Cloud continuum. This is especially prominent when resource fragmentation –where available resources become isolated and underutilized– becomes an issue due to the frequent reallocations of VNFs. However, traditional optimization methods often struggle to deal with the dynamic and complex nature of the VNF placement problem when fragmentation is considered. This study proposes a novel online VNF placement approach for Edge/Cloud infrastructures that utilizes Deep Reinforcement Learning (DRL) and Reward Constrained Policy Optimization (RCPO) to address this problem. We combine DRL’s adaptability with RCPO’s constraint incorporation capabilities to ensure that the learned policies satisfy the performance and resource constraints while minimizing resource fragmentation. Specifically, the VNF placement problem is first formulated as an offline-constrained optimization problem, and then we devise an online solver using Neural Combinatorial Optimization (NCO). Our method incorporates a metric called Resource Fragmentation Degree (RFD) to quantify fragmentation in the network. Using this metric and RCPO, our NCO agent is trained to make intelligent placement decisions that reduce fragmentation and optimize resource utilization. An error correction heuristic complements the robustness of the proposed framework. Through extensive testing in a simulated environment, the proposed approach is shown to outperform state-of-the-art VNF placement techniques when it comes to minimizing resource fragmentation under constraint satisfaction guarantees.

Low Earth Orbit Satellite Networks (LSNs) are integral to supporting a broad range of modern applications, which are typically modeled as Service Function Chains (SFCs). Each SFC is composed of Virtual Network Functions (VNFs), where each VNF performs a specific task. In this work, we tackle two key challenges in deploying SFCs across an LSN. Firstly, we aim to optimize the long-term system performance by minimizing the average end-to-end SFC execution delay, given that each satellite comes with a pre-installed/cached subset of VNFs. To achieve optimal SFC placement, we formulate an offline Dynamic Programming (DP) equation. To overcome the challenges associated with DP, such as its complexity, the need for probability knowledge, and centralized decision-making, we put forth an online Multi-Agent Q-Learning (MAQL) solution. Our MAQL approach addresses convergence issues in the non-stationary LSN environment by enabling satellites to share learning parameters and update their Q-tables based on distinct rules for their selected actions. Secondly, to determine the optimal VNF subsets for satellite caching, we develop a Bayesian Optimization (BO)-based learning mechanism that operates both offline and continuously in the background during runtime. Extensive experiments demonstrate that our MAQL approach achieves near-optimal performance comparable to the DP model and significantly outperforms existing baselines. Moreover, the BO-based approach effectively enhances the request serving rate over time.