The joint occupation of urban road infrastructures by logistic flows and general mobility traffic, typical of contexts in which a port is located within a city, is a known source of unwanted externalities such as congestion and pollution. In this paper we propose a model-predictive control scheme aimed at maximizing the throughput of vehicles into the zone where the two kinds of traffic interact, thus mitigating the impact of negative phenomena, as well as increasing the overall economic benefit of the port-city system. In order to provide the detailed representation of the network dynamics required by the predictive controller, we define a model supported by micro-simulation of the port-city area. Then, since the direct application of micro-simulation is not feasibile in real-time, we resort to a surrogate data-driven approximation to be used by the controller. To this purpose, we define a principled sampling design scheme aimed at yielding a suitable training set for the surrogate model, and provide a theoretical analysis on the conditions to guarantee an efficient covering of the relevant sets. A simulation case study involving the port-city context of Genova in north-west Italy is presented, in order to showcase the advantages of the proposed sampling scheme and the performance of the model-predictive controller as well.

Transformer-based large language models (LLMs) have achieved remarkable success across natural language processing tasks, and their abilities have been exploited also for dynamic system control, particularly for robots. However, their practical deployment faces critical limitations: reliance on remote servers, substantial computational requirements, and high energy consumption make their practical application to real-time control very challenging. Small language models (SLMs), characterized by significantly fewer parameters, offer a promising alternative by enabling local execution, preserving privacy, and facilitating customization. Despite these advantages, SLMs remain underexplored for dynamic system control. This paper investigates whether SLMs can effectively control dynamic systems in real-time within a model-predictive control (MPC) framework. We propose a novel approach that leverages natural language prompting and a self-assessment strategy to generate effective cost functions for MPC, ensuring real-time adaptability and robustness. Unlike typical existing reward-shaping methods, our methodology is designed to be applicable to arbitrary systems and goals. Through extensive simulations, we test the methodology under diverse dynamic systems and tasks, validating the feasibility of SLM-based real-time control.

Defining effective policies for managing traffic in large cities, particularly near major logistics hubs such as ports, is a challenging problem due to the critical interaction between mobility and freight flows. In this work, we combine a traffic simulator with a change-detection test to assess a priori whether a specific policy would have an impact on city mobility. More specifically, we propose a general methodology to identify the expected number of days/monitoring samples before gaining evidence that the policy has introduced a detectable change in the traffic data acquired after enforcing the policy. Our experiments, conducted on simulated traffic, focused on the port-city context of Genova, showcase that our proposed methodology can provide outcomes that are consistent with the kind and expected effectiveness of policies under evaluation.

The need to find solutions in function form that optimize a given nonlinear cost functional arises routinely in many important areas of operations research and applied mathematics. In most practical cases, problems of this kind require a numerical solution based on some suitable class of approximating architectures. This paper introduces the use of binary Voronoi linear trees (BVLTs) for the approximate solution of a general class of functional optimization problems. The main features of the considered trees are (i) a splitting scheme based on a Voronoi bisection criterion and (ii) linear outputs in the leaves, which make the resulting models more flexible compared to classic trees with cuts parallel to the axes and constant outputs. At the same time, due to the binary recursive structure, BVLTs retain the well-known efficiency of decision tree architectures. Consistently with the typical tree construction framework, we provide a greedy algorithm for the approximate solution of the addressed functional optimization problem. Universal approximation capabilities of the proposed class of models are derived in the theoretical analysis, and the consistency of the solution is discussed as well. In order to improve accuracy and robustness, we also consider the use of BVLTs in ensemble fashion, through an aggregation scheme well suited to optimization purposes. Simulation tests involving various optimization problems are presented, showing how the proposed algorithm can cope well in complex multivariate contexts, especially in ensemble form.

Approximate dynamic programming (ADP) is the standard technique to derive optimal policies in finite-horizon stochastic multistage optimal decision problems, with continuous state space. Yet, it presents two main issues. First, it requires several nonlinear optimizations in state sample points, in order to generate training sets for value function and policy approximations. Then, the aforementioned approximations are obtained in pure pattern/target fashion, in a blind way with respect to system performance and possible constraints. In this paper we show how a special deep learning structure, coupled with a loss function that takes into account both Bellman’s equation and the approximation of the value function, enables a more efficient implementation of the ADP framework. In particular, the proposed solution boils down to a single training of the deep network parameters, eliminating the need for the pointwise minimizations and approximations of ADP. The computation of the policies explicitly takes into account cost performance and constraints as well. In order to enhance the efficiency of the procedure, low-discrepancy sampling is considered for the state points sampling. A theoretical analysis is provided to ensure the correctness and consistency of the method. Then, simulation results are provided to showcase the method both in a finite-horizon and a receding-horizon setting.

In this work we investigate the use of ensemble methods, consisting in the aggregation of several approximating models, in the context of functional optimization. In fact, while ensemble techniques are routinely employed in the machine learning literature for classification and regression, there is little research on their application to general optimization problems. Here we consider two strategies to aggregate different solutions to a functional optimization problem, based on optimized weighted averaging and aggregation over the minimum, the latter also in approximate version. A theoretical analysis of approximate functional optimization in the context of ensemble aggregation is provided. Then, simulation results are reported to showcase the advantages of ensembles for functional optimization, in terms of better accuracy and improved robustness with respect to single solutions.

In this paper, we present a method, based on model predictive control (MPC), to reduce the impact of pollutant emissions in contexts where a port is located within a city. To this purpose, we first introduce a dynamic model of the interactions between truck flows generated by the port and general mobility traffic in the shared urban infrastructure at the port-city interface. In order to keep track of the multiclass and complex nature of the system, the model takes advantage of microsimulation and deep learning for the prediction of road network traffic and related pollutant emissions. Then, we define a MPC control scheme exploiting the proposed model, to be used in real time to maintain the emissions levels below a certain threshold by appropriately adjusting traffic inflows from the port to the city, which represent the controls optimized by the MPC procedure. A simulation case study, involving the port of Genova in north-west Italy, is presented to showcase the ability of the proposed MPC scheme to control emissions in the shared area, also in complex situations such as transitions to mobility rush hours.

We investigate the use of traffic simulation tools combined with surrogate models based on neural networks and interpretable machine learning techniques, to analyze and explain complex dynamics at play in urban traffic contexts. We focus in particular on signalized junctions, as one of the main elements affecting the performance of an urban traffic network. Evaluating the impact of traffic light programs and traffic flows can provide very useful information that can be exploited for monitoring and optimization purposes. Yet, a detailed and systematic evaluation is difficult when the area of interest is not trivial, especially when time-consuming micro-simulation tools are employed. In the paper we show how interpretable machine learning techniques, in combination with traffic micro-simulation tools, can provide a computationally efficient way to evaluate the impact of a network of signalized junctions in a given zone. Simulation tests are presented to showcase the methodology in a blueprint scenario, related to the interactions between traffic dynamics and traffic lights behavior.

We propose a model predictive control (MPC) scheme for the real-time optimization of interactions between logistic traffic and urban mobility, in contexts where a port is located within a city center. The aim is to mitigate negative phenomena related to traffic sharing the same urban space, through the efficient operation of existing infrastructure at a tactical/operational level. To this purpose, we first define an integrated model of the port/city interactions, focusing on the import and export chains in proximity of the port where trucks and general mobility vehicles interact. Then, we implement an MPC approach in order to optimize an objective function reflecting the interests of the different involved stakeholders. An important feature of the presented model is that it allows the analysis of different levels of cooperation between the city and the port, highlighting possible advantages of increased coordination at the operational level. To evaluate the effect of the proposed control scheme, we present a case study based on the port of Genova in north-west Italy.

We introduce a receding-horizon dynamic optimization approach for the real-time efficient management of conflicting traffic flows from a port terminal and urban mobility, in contexts where the port is located within a city center. This common situation typically entails a negative impact on the urban transport network, where the need to share the infrastructure causes issues both to the logistic and mobility flows. For the purpose of mitigating the negative impact, we formalize a model-predictive control scheme based on an integrated model of the port-city environment, focusing on the import and export chains in proximity of the terminal where trucks and general mobility vehicles interact. By exploiting available forecasts on mobility demand and container traffic, together with the coordinated action of perimetral and gate control, our proposed scheme aims at optimizing a desired performance index reflecting the interests of the involved stakeholders. Simulation results are presented regarding a context based on the port of Genova in north-west Italy, showcasing the ability of the proposed receding-horizon scheme to manage complex situations such as bottlenecks and transitions to mobility rush hours.

We propose an algorithm for the approximate solution of general nonlinear functional optimization problems through recursive binary Voronoi tree models. Unlike typical binary tree structures commonly employed for classification and regression problems, where splits are performed parallel to the coordinate axes, here the splits are based on Voronoi cells defined by a pair of centroids. Models of this kind are particularly suited to functional optimization, where the optimal solution function can easily be discontinuous even for very smooth cost functionals. In fact, the flexible nature of Voronoi recursive trees allows the model to adapt very well to possible discontinuities. In order to improve efficiency, accuracy and robustness, the proposed algorithm exploits randomization and the ensemble paradigm. To this purpose, an ad hoc aggregation scheme is proposed. Simulation tests involving various test problems, including the optimal control of a crane-like system, are presented, showing how the proposed algorithm can cope well with discontinuous optimal solutions and outperform trees based on the standard split scheme.

One the most attractive features of urban traffic network models is the possibility of running hypothetical scenarios to evaluate the impact of strategic and tactical decisions. In order to provide statistically meaningful results, the simulation runs should be able to capture the complex multivariate distributions characterizing the involved variables, and a key factor is the correct modeling of possible statistical dependence among the generated inputs used to define the desired scenarios. Here we introduce a data-driven method for scenario generation based on the statistical concept of copula models, through which the marginals of single input parameters can be chosen freely without altering the joint multivariate dependence structure of the inputs. This approach is particularly suited to running what-if scenarios, in which the marginal distributions of the inputs are changed, while retaining the general joint dependence scheme. The method exploits only a finite set of measures from the network and copes with arbitrary sets of input parameters without requiring any assumption on the kind of traffic model or the shape of the involved multivariate distributions. Simulation tests involving different scenarios show that the proposed method is able to capture complex multivariate distributions of the simulation outcomes and yield reliable inferences in what-if analyses, significantly better than in the case the joint dependence is ignored.

The calibration of urban traffic microsimulation models is addressed in this paper, focusing on the multivariate distribution of traffic features. This allows to take into account day-to-day variability and possibly complex statistical dependencies, enabling robust validation and the computation of more accurate statistics. Variability in traffic models can be obtained, in principle, by suitably tuning origin-destination demand flows, but this becomes problematic for microsimulation models when complex multivariate distributions are considered, due to excessive dimensionality and nonlinearity. Thus, here we investigate how complex multivariate behavior can be achieved by suitably varying only a subset of the parameters ruling the microsimulation dynamics, without imposing any distribution on the demand flows. To this purpose, rather than looking for fixed values of the parameters, we exploit distribution models having the structure of mixtures of uniforms, through which the simulator parameter values can be randomly sampled, thus replacing the simulator inner randomization mechanism. We formalize the optimization of the mixture parameters through a maximum mean discrepancy principle involving the simulator dynamics, which leads to a large-dimensional non-differentiable problem that we solve through a method based on cross-entropy. Preliminary experiments concerning the application of the proposed methodology to the SUMO simulator show how it is possible to capture quite complex multivariate distributions of target flows varying only four parameters.

In this paper we investigate the use of nonlinear embeddings to represent dynamic inputs in surrogate models, used to estimate the outcome of time-consuming simulation tools in feasible time. By encoding the temporal sequences, together with non-dynamic inputs, into a space of static features, we are able to formalize the surrogate modeling problem as a standard supervised learning one. To this purpose, we propose to use the echo state network (ESN) paradigm to generate the embeddings of the dynamic inputs. The main advantage of this approach is that the embedding does not require training, since it is provided by a reservoir of neurons with randomly generated parameters. In order to enhance the robustness of this method based on randomization, we propose to use an ensemble of different ESN embeddings. Within this scheme, we consider a procedure aimed at controlling diversity among the ensemble elements before the simulations are run. Furthermore, in order to improve accuracy, we investigate also nonlinear mappings for the generation of the ESN outputs. The proposed approach is applied to an urban traffic network example, a typical case in which simulation models are generally complex and very time-consuming, and the scenarios to be simulated involve both static inputs and time-varying quantities.

We have developed an imitation learning approach for the image-based control of a low-cost low-accuracy robot arm. The image-based control of manipulation arms is still an unsolved problem, at least under challenging conditions such as those here addressed. Many attempts for solutions in the literature are based on machine learning, generally relying on deep neural network architectures. In typical imitation approaches, the deep network learns from a human expert. In our case the network is trained on state/action pairs obtained through a Belief Space Planning algorithm, a stochastic method that requires only a rough tuning, particularly suited to unstructured and dynamic environments. Our approach allows to obtain a lightweight manipulation system that demonstrated its efficiency, robustness and good performance in real-world tests, and that is reproducible in experiments and results, despite its inaccuracy and non-repeatable kinematics. The proposed system performs well on a simple reaching task, requiring limited training on our quite challenging platform. The main contribution of the proposed work lies in the definition and real-world testing of an efficient controller, based on the integration of Belief Space Planning with the imitation learning paradigm, that enables even inaccurate, very low-cost robotic manipulators to be actually controlled and employed in the field.

This paper investigates the policy optimization paradigm, where a learning model is trained to find the solution of complex Markov decision problems, as a tool to address the berth allocation problem in multimodal terminals. To this purpose, we drop the typical formulation of the latter as a mixed-integer static scheduling one, and we model it instead as an evolving scenario in which berths are assigned to ships according to a parameterized policy function that drives the temporal evolution of the environment. We adopt a cross-entropy optimization scheme to optimize the policy parameters, which is a simple and highly parallelizable gradient-free technique. As compared to the static mixed-integer formulation, the proposed approach relies on a much lighter optimization problem in the continuous space of the policy parameters, thus making it feasible to replan in real time when needed. Furthermore, the generality of the policy optimization approach allows to take into account any performance metric and specific feature of the scenario straightforwardly, without the need to devise ad hoc heuristics. Simulation tests showcase the good performance of the policy approach under various conditions.

In this paper we investigate deep learning architectures combined with low-discrepancy sampling as surrogate models. The aim is to provide a quick estimate of the outcome of an expensive process when many evaluations are needed, e.g., for optimization purposes. The simulation of urban traffic, routinely employed for the design and performance evaluation of policies in urban mobility scenarios, is an example of procedure characterized by very complex dynamics, large dimensionality and intensive computational requirements. In this context, the need for a surrogate model arises in many forms, e.g., for origin-destination demand calibration, traffic light optimization, strategic planning. In order to cope with this complexity and large dimensionality we take advantage of the excellent approximating capabilities of deep neural networks as learning models. Then, we employ low-discrepancy sequences as sampling designs for the simulation runs required to create the training set for the deep surrogate model. This kind of sampling guarantees deterministically a good covering of the input space, and is able to exploit possible regularities of the simulation outcome. Extensive experimental tests are presented involving the popular SUMO microsimulator, showing the advantages of the proposed surrogate modeling solution under various performance measures.

We propose a method based on recursive binary Voronoi trees to learn a nonparametric model of the distribution underlying a given dataset. The obtained model can be used as a general tool both to extract good samples from the original dataset (e.g., for batch selection, bagging, or sample size reduction) or to generate new synthetic ones, also in a conditional fashion (e.g., to deal with imbalanced sets or to reconstruct corrupted points). In order to ensure that the distribution of the new sets, either sampled or generated, follows closely that of the original dataset, we design all the procedures according to a specific measure of distance between distributions. The use of binary recursive Voronoi structures enables the proposed algorithms to be simple, efficient and able to adapt to the shape of the original dataset. Simulation tests showcase the good performance and flexibility of the approach in various learning contexts. (C) 2019 Elsevier Ltd. All rights reserved.

Off-line supervised learning from data of robustly-stabilizing nonlinear explicit model predictive controllers (EMPC) is dealt with in this letter. The learning procedure relies on the construction of a suitably large set of specifically chosen sampling points of the state space in which the values of the optimal EMPC control function have to be computed. When bounding the magnitude of approximation errors is important for stability or performance specifications, regular gridding techniques are not feasible due to the curse of dimensionality arising from the structural exponential growth of the number of points with the state dimension. In this note, we consider non-regular sampling techniques - namely, i.i.d. sampling with uniform distribution, low-discrepancy sequences and lattice point sets - that offer a good covering of the state space without suffering from an unfeasible growth of the number of points, while preserving at the same time the method guarantees in terms of robustness and stability. Some theoretical properties of the proposed sampling schemes are briefly discussed, and their successful application is showcased in a practically-relevant optimal heating problem involving a 21-dimensional state space that rules out the use of regular gridding techniques.

A novel measurement approach for power-flow analysis in medium-voltage (MV) networks, based on load power measurements at low-voltage level in each secondary substation (SS) and only one voltage measurement at the MV level at primary substation busbars, was proposed by the authors in previous works. In this paper, the method is improved to cover the case of temporary unavailability of load power measurements in some SSs. In particular, a new load power estimation method based on artificial neural networks (ANNs) is proposed. The method uses historical data to train the ANNs and the real-time available measurements to obtain the load estimations. The load-flow algorithm is applied with the estimated load powers, and the MV network state variables are obtained. The proposed method is validated for the real MV distribution network of the island of Ustica. The loads of selected SSs are estimated for two full days of different seasons. In comparison with previous works, satisfactory results are obtained in terms of uncertainty in the calculated power flows, thus suggesting the applicability of the proposed method for real-time monitoring of MV distribution networks.