| 1. |
- Varga, Balázs, 1990, et al.
(författare)
-
Robust tracking controller design for active dolly steering
- 2018
-
Ingår i: Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering. - : SAGE Publications. - 2041-2991 .- 0954-4070. ; 232:5, s. 695-706
-
Forskningsöversikt (refereegranskat)abstract
- In this paper different actuation level steering control methods for an A-double vehicle combination (tractor-semitrailer-dolly-semitrailer) are proposed. The aim of the paper is to show viability of advanced actuation control strategies on a practical vehicular application. Three different types of robust controllers are proposed: a robust Proportional Integral Derivative (PID) controller, an output feedback linear Hinf controller and an induced L2-norm minimizing Linear Parameter Varying (LPV) controller. All controllers are augmented with anti-windup compensators to respect steering angle and steering rate limits. Each model based controller robustly rejects external disturbances and tracks a reference steering angle, generated by motion control system. Frequency- and time domain analysis proves that Hinf and LPV controllers outperform PID controller in terms of reference tracking and disturbance rejection. Comparative simulation scenarios are provided on the basis of Volvo Group Trucks Technology’s high fidelity vehicle simulator.
|
|
| 2. |
- Andersson, Viktor, 1995, et al.
(författare)
-
Controlled Decent Training
- 2023
-
Tidskriftsartikel (övrigt vetenskapligt/konstnärligt)abstract
- In this work, a novel and model-based artificial neural network (ANN) training method is developed supported by optimal control theory. The method augments training labels in order to robustly guarantee training loss convergence and improve training convergence rate. Dynamic label augmentation is proposed within the framework of gradient descent training where the convergence of training loss is controlled. First, we capture the training behavior with the help of empirical Neural Tangent Kernels (NTK) and borrow tools from systems and control theory to analyze both the local and global training dynamics (e.g. stability, reachability). Second, we propose to dynamically alter the gradient descent training mechanism via fictitious labels as control inputs and an optimal state feedback policy. In this way, we enforce locally H2 optimal and convergent training behavior. The novel algorithm, Controlled Descent Training (CDT), guarantees local convergence. CDT unleashes new potentials in the analysis, interpretation, and design of ANN architectures. The applicability of the method is demonstrated on standard regression and classification problems.
|
|
| 3. |
- Varga, Balázs, 1990, et al.
(författare)
-
Constrained Policy Gradient Method for Safe and Fast Reinforcement Learning: a Neural Tangent Kernel Based Approach
- 2021
-
Tidskriftsartikel (övrigt vetenskapligt/konstnärligt)abstract
- This paper presents a constrained policy gradient algorithm. We introduce constraints for safe learning with the following steps. First, learning is slowed down (lazy learning) so that the episodic policy change can be computed with the help of the policy gradient theorem and the neural tangent kernel. Then, this enables us the evaluation of the policy at arbitrary states too. In the same spirit, learning can be guided, ensuring safety via augmenting episode batches with states where the desired action probabilities are prescribed. Finally, exogenous discounted sum of future rewards (returns) can be computed at these specific state-action pairs such that the policy network satisfies constraints. Computing the returns is based on solving a system of linear equations (equality constraints) or a constrained quadratic program (inequality constraints). Simulation results suggest that adding constraints (external information) to the learning can improve learning in terms of speed and safety reasonably if constraints are appropriately selected. The efficiency of the constrained learning was demonstrated with a shallow and wide ReLU network in the Cartpole and Lunar Lander OpenAI gym environments. The main novelty of the paper is giving a practical use of the neural tangent kernel in reinforcement learning.
|
|
| 4. |
- Varga, Balázs, 1990, et al.
(författare)
-
Data-driven distance metrics for kriging - Short-term urban traffic state prediction
- 2023
-
Ingår i: IEEE Transactions on Intelligent Transportation Systems. - 1524-9050 .- 1558-0016. ; 24:6, s. 6268-6279
-
Tidskriftsartikel (refereegranskat)abstract
- Estimating traffic flow states at unmeasured urban locations provides a cost-efficient solution for many ITS applications. In this work, a geostatistical framework, kriging is extended in such a way that it can both estimate and predict traffic volume and speed at various unobserved locations, in real-time. In the paper, different distance metrics for kriging are evaluated. Then, a new, data-driven one is formulated, capturing the similarity of measurement sites. Then, with multidimensional scaling the distances are transformed into a hyperspace, where the kriging algorithm can be used. As a next step, temporal dependency is injected into the estimator via extending the hyperspace with an extra dimension, enabling for short horizon traffic flow prediction. Additionally, a temporal correction is proposed to compensate for minor changes in traffic flow patterns. Numerical results suggest that the spatio-temporal prediction can make more accurate predictions compared to other distance metric-based kriging algorithms. Additionally, compared to deep learning, the results are on par while the algorithm is more resilient against traffic pattern changes.
|
|
| 5. |
- Varga, Balázs, 1990, et al.
(författare)
-
Deep Q-learning: a robust control approach
- 2023
-
Ingår i: International Journal of Robust and Nonlinear Control. - : Wiley. - 1099-1239 .- 1049-8923. ; 33:1, s. 526-544
-
Tidskriftsartikel (refereegranskat)abstract
- This work aims at constructing a bridge between robust control theory and reinforcement learning. Although, reinforcement learning has shown admirable results in complex control tasks, the agent’s learning behaviour is opaque. Meanwhile, system theory has several tools for analyzing and controlling dynamical systems. This paper places deep Q-learning is into a control-oriented perspective to study its learning dynamics with well-established techniques from robust control. An uncertain linear time-invariant model is formulated by means of the neural tangent kernel to describe learning. This novel approach allows giving conditions for stability (convergence) of the learning and enables the analysis of the agent’s behaviour in frequency-domain. The control-oriented approach makes it possible to formulate robust controllers that inject dynamical rewards as control input in the loss function to achieve better convergence properties. Three output-feedback controllers are synthesized: gain scheduling H2, dynamical Hinf, and fixed-structure Hinf controllers. Compared to traditional deep Q-learning techniques, which involve several heuristics, setting up the learning agent with a control-oriented tuning methodology is more transparent and has well-established literature. The proposed approach does not use a target network and randomized replay memory. The role of the target network is overtaken by the control input, which also exploits the temporal dependency of samples (opposed to a randomized memory buffer). Numerical simulations in different OpenAI Gym environments suggest that the Hinf controlled learning can converge faster and receive higher scores (depending on the environment) compared to the benchmark Double deep Q-learning.
|
|
| 6. |
- Varga, Balázs, 1990, et al.
(författare)
-
Energy-aware predictive control for electrified bus networks
- 2019
-
Ingår i: Applied Energy. - : Elsevier BV. - 1872-9118 .- 0306-2619. ; 252
-
Tidskriftsartikel (refereegranskat)abstract
- For an urban bus network to operate efficiently, conflicting objectives have to be considered: providing sufficient service quality while keeping energy consumption low. The paper focuses on energy efficient operation of bus lines, where bus stops are densely placed, and buses operate frequently with possibility of bunching. The proposed decentralized, bus fleet control solution aims to combine four conflicting goals incorporated into a multi-objective, nonlinear cost function. The multi-objective optimization is solved under a receding horizon model predictive framework. The four conflicting objectives are as follows. One is ensuring periodicity of headways by watching leading and following vehicles i.e. eliminating bus bunching. Equal headways are only a necessary condition for keeping a static, predefifined, periodic timetable. The second objective is timetable tracking, and passenger waiting time minimization. In case of high passenger demand, it is desirable to haste the bus in order to prevent bunching. The final objective is energy efficiency. To this end, an energy consumption model is formulated considering battery electric vehicles with recuperation during braking. Alternative weighting strategies are compared and evaluated through realistic scenarios, in a calibrated microscopic traffic simulation environment. Simulation results confirm of 3-8% network level energy saving compared to bus holding control while maintaining punctuality and periodicity of buses.
|
|
| 7. |
- Varga, Balázs, 1990, et al.
(författare)
-
Optimal headway merging for balanced public transport service in urban networks
- 2018
-
Ingår i: IFAC-PapersOnLine. - : Elsevier BV. - 2405-8963. ; 51:9, s. 416-421
-
Konferensbidrag (refereegranskat)abstract
- This paper presents a velocity control/advise algorithm relying on vehicle-to-vehicle communication, to ensure the headway homogeneity of buses on a joint corridor, i.e. when multiple lines merge and travel on the same route. The proposed control method first schedules merging buses prior to entering a common line. Second, based on the position and velocity of the bus ahead of the controlled one, a shrinking horizon model predictive controller (MPC) calculates a proper velocity profile for the merging bus. The model is able to predict short time- space behavior of public transport buses enabling constrained, finite horizon, optimal control solution to reach the merging point with equidistant headways, taking all buses from different lines into account. The controller is tested in a high fidelity traffic simulator with realistic scenarios.
|
|
| 8. |
- Varga, Balázs, 1990, et al.
(författare)
-
Optimally combined headway and timetable reliable public transport system
- 2018
-
Ingår i: Transportation Research, Part C: Emerging Technologies. - : Elsevier BV. - 0968-090X. ; 92, s. 1-26
-
Tidskriftsartikel (refereegranskat)abstract
- This paper presents a model-based multiobjective control strategy to reduce bus bunching and hence improve public transport reliability. Our goal is twofold. First, we define a proper model, consisting of multiple static and dynamic components. Bus-following model captures the longitudinal dynamics taking into account the interaction with the surrounding traffic. Furthermore, bus stop operations are modeled to estimate dwell time. Second, a shrinking horizon model predictive controller (MPC) is proposed for solving bus bunching problems. The model is able to predict short time-space behavior of public transport buses enabling constrained, finite horizon, optimal control solution to ensure homogeneity of service both in time and space. In this line, the goal with the selected rolling horizon control scheme is to choose a proper velocity profile for the public transport bus such that it keeps both timetable schedule and a desired headway from the bus in front of it (leading bus). The control strategy predicts the arrival time at a bus stop using a passenger arrival and dwell time model. In this vein, the receding horizon model predictive controller calculates an optimal velocity profile based on its current position and desired arrival time. Four different weighting strategies are proposed to test (i) timetable only, (ii) headway only, (iii) balanced timetable - headway tracking and (iv) adaptive control with varying weights. The controller is tested in a high fidelity traffic simulator with realistic scenarios. The behavior of the system is analyzed by considering extreme disturbances. Finally, the existence of a Pareto front between these two objectives is also demonstrated.
|
|
