publications

2025

1. Master Thesis

   The Price of Constraints in Linear Tracking Model Predictive Control

   Anastasios Vlachos

   ETH Zurich, 2025

   Abs PDF

   Trajectory tracking constitutes a core challenge in control systems, applicable across various domains like robotics and autonomous vehicles. While unconstrained tracking is well-understood and solvable using classical methods, introducing constraints fundamentally alters the landscape of achievable performance. Constrained tracking problems are prevalent in real-world applications, where safety and operational limits must be respected. This thesis investigates the performance degradation, or suboptimality, introduced by input constraints in tracking control, using Model Predictive Control (MPC) as the framework of study. We focus on general time-varying references and derive complexity bounds that quantify the cost of constraints, relative to the unconstrained optimal solution. Our analysis applies to linear systems with polytopic input constraints and introduces a two-step transformation to ensure reference feasibility. We extend known results on the structure properties of MPC policies to the tracking setting and provide theoretical guarantees on the suboptimality gap. Numerical simulations corroborate our theoretical findings and demonstrate the practical relevance of our approach. This work offers new insights into the trade-offs inherent in constrained tracking and sets the stage for future developments in online control scenarios.

2024

1. CDC 2024

   Online Residual Learning from Offline Experts for Pedestrian Tracking

   Anastasios Vlachos, Anastasios Tsiamis, Aren Karapetyan, and 2 more authors

   In 2024 IEEE 63rd Conference on Decision and Control (CDC), 2024

   Abs PDF

   In this paper, we consider the problem of predicting unknown targets from data. We propose Online Residual Learning (ORL), a method that combines online adaptation with offline-trained predictions. At a lower level, we employ multiple offline predictions generated before or at the beginning of the prediction horizon. We augment every offline prediction by learning their respective residual error concerning the true target state online, using the recursive least squares algorithm. At a higher level, we treat the augmented lower-level predictors as experts, adopting the Prediction with Expert Advice framework. We utilize an adaptive softmax weighting scheme to form an aggregate prediction and provide guarantees for ORL in terms of regret. We employ ORL to boost performance in the setting of online pedestrian trajectory prediction. Based on data from the Stanford Drone Dataset, we show that ORL can demonstrate best-of-both-worlds performance.

2021

1. MOCAST 2021

   Exploring the Effectiveness of Sigma-Delta Modulators in Stochastic Computing-Based FIR Filtering

   Anastasios Vlachos, Nikos Temenos, and Paul P. Sotiriadis

   In 2021 10th International Conference on Modern Circuits and Systems Technologies (MOCAST), 2021

   Abs PDF

   A soft-filtering processing architecture based on Sigma-Delta Modulation and Stochastic Computing is proposed. It converts a high-resolution signal using a first order digital Sigma-Delta Modulator into a single-bit one and then exploits Stochastic Computing’s encoding to perform area-efficient multiplications. The Sigma-Delta Modulator allows for the input signal to be oversampled at a much higher frequency rate, offering improved performance in terms of SNR, which is not possible with standard Stochastic Computing filter realizations. Spectral simulations results demonstrate the proper signal quantization and operation of the filter, including the filter’s roll-off behavior. FPGA synthesis results of the proposed architecture, illustrate its area advantages in comparison to conventional binary filtering.