We study the turnpike phenomenon for optimal control problems with mean-field dynamics that are obtained as the limit $N\rightarrow \infty$ of systems governed by a large number $N$ of ordinary differential equations. We show that the optimal control problems with large time horizons give rise to a turnpike structure of the optimal state and the optimal control. For the proof, we use the fact that the turnpike structure for the problems on the level of ordinary differential equations is preserved under the corresponding mean-field limit.

The connection between Residual Neural Networks (ResNets) and continuous-time control systems (known as NeurODEs) has led to a mathematical analysis of neural networks, which has provided interesting results of both theoretical and practical significance. However, by construction, NeurODEs have been limited to describing constant-width layers, making them unsuitable for modelling deep learning architectures with layers of variable width. In this paper, we propose a continuous-time Autoencoder, which we call AutoencODE, based on a modification of the controlled field that drives the dynamics. This adaptation enables the extension of the mean-field control framework originally devised for conventional NeurODEs. In this setting, we tackle the case of low Tikhonov regularisation, resulting in potentially non-convex cost landscapes. While the global results obtained for high Tikhonov regularisation may not hold globally, we show that many of them can be recovered in regions where the loss function is locally convex. Inspired by our theoretical findings, we develop a training method tailored to this specific type of Autoencoders with residual connections, and we validate our approach through numerical experiments conducted on various examples.

Compliant interaction between robots and the environment is crucial for completing contact-rich tasks. However, obtaining and implementing optimal interaction behavior in complex unknown environments remains a challenge. This article develops a hybrid impedance and admittance control (HIAC) scheme for robots subjected to the second-order unknown environment. To obtain the second-order target impedance model that represents the optimal interaction behavior without the accurate environment dynamics and acceleration feedback, an impedance adaptation method with virtual inertia is proposed. Since impedance control and admittance control have complementary structures and result in unsatisfactory performance in a wide range of environmental stiffness due to their fixed causality, a hybrid system framework suitable for the second-order environment is proposed to generate a series of intermediate controllers which interpolate between the responses of impedance and admittance controls by using a switching controller and adjusting its switching duty cycle. In addition, the optimal intermediate controller is selected using a mapping of the optimal duty cycle to provide the optimal implementation performance for the target impedance model. The proposed HIAC scheme can achieve the desired interaction and impedance implementation performance while ensuring system stability. Simulation and experimental studies are performed to verify the effectiveness of our scheme with a 2-DOF manipulator and a 7-DOF Franka EMIKA panda robot, respectively.

This paper proposes a linear quadratic approximation approach to dynamic nonlinear rationally inattentive control problems with multiple states and multiple controls. An efficient toolbox to implement this approach is provided. Applying this toolbox to five economic examples demonstrates that rational inattention can help explain the comovement puzzle in the macroeconomics literature.

We consider the problem of controlling the drift and diffusion rate of the endowment processes of two firms such that the joint survival probability is maximized. We assume that the endowment processes are continuous diffusions, driven by independent Brownian motions, and that the aggregate endowment is a Brownian motion with constant drift and diffusion rate. Our results reveal that the maximal joint survival probability depends only on the aggregate risk-adjusted return and on the maximal risk-adjusted return that can be implemented in each firm. Here the risk-adjusted return is understood as the drift rate divided by the squared diffusion rate.

Wearable robots, sometimes known as exoskeletons, are incredible devices for improving human strength, reducing fatigue, and restoring impaired mobility. The control of powered exoskeletons, on the other hand, is still a challenge. This necessitates the development of a technique to simulate exoskeleton–wearer interaction. This study uses a two-dimensional human skeletal model with a powered knee exoskeleton to predict the optimal lifting motion and assistive torque. For lifting motion prediction, an inverse dynamics optimization formulation is utilized. In addition, the electromechanical dynamics of the exoskeleton DC motor are modeled in the lifting optimization formulation. The design variables are human joint angle profiles and exoskeleton motor current profiles. The human joint torque square is minimized subject to physical and lifting task constraints. Then, the lifting optimization problem is solved by the gradient-based sparse nonlinear optimizer (SNOPT). Furthermore, the optimal exoskeleton torque is implemented through a two-phase control strategy to provide optimal assistance in lifting. Experimental validations of the optimal control with 6 Nm and 16 Nm maximum assistive torque are presented. Both 6 Nm and 16 Nm maximum optimal assistance of the exoskeletons reduce the mean values of vastus lateralis, biceps femoris, and latissimus dorsi muscle activations compared to the lifting without the exoskeleton. However, the mean value of the vastus medialis activation is increased by a small amount for the exoskeleton case, although its peak value is reduced. Finally, the experimental results demonstrate that the proposed lifting optimization formulation and control strategy are promising for powered knee exoskeleton for lifting tasks.

A robotic system constructed as a wheeled inverted pendulum (WIP) serves as an impressive demonstrator, since this kind of system is inherently nonlinear, unstable, and nonminimum phase. These properties may pose several difficulties, when it comes to control and trajectory planning. This paper shows a method for deriving a highly dynamic trajectory compliant with the system dynamics by means of solving an optimal control problem (OCP) using multiple shooting. The assumed task includes that the WIP should pass a height-restricting barrier. This can be achieved by leaning back or forth, in order to reduce the overall height of the WIP. The constraints inherent to the definition of this trajectory are nonconvex due to the shape of the robot. The constraint functions have a local minimum in an infeasible region. This can cause problems when the initial guess is within this infeasible region. To overcome this, a multistage approach is proposed for this special OCP to evade the infeasible local minimum. After solving four stages of subsequent optimization problems, the optimal trajectory is obtained and can be used as feedforward for the real system.

For greater autonomy of visual control-based solutions, especially applied to mobile robots, it is necessary to consider the existence of unevenness in the navigation surface, an intrinsic characteristic of several real applications. In general, depth information is essential for navigating three-dimensional environments and for the consistent parameter calibration of the visual models. This work proposes a new solution, including depth information in the visual path-following (VPF) problem, which allows the variation of the perception horizon at runtime while forcing the coupling between optical and geometric quantities. A new NMPC (nonlinear model predictive control) framework considering the addition of a new input to an original solution for the constrained VPF-NMPC allows the maintenance of low computational complexity. Experimental results in an outdoor environment with a medium-sized commercial robot demonstrate the correctness of the proposal.

In this paper, the turnpike property is established for a nonconvex optimal control problem in discrete time. The functional is defined by the notion of the ideal convergence and can be considered as an analogue of the terminal functional defined over infinite-time horizon. The turnpike property states that every optimal solution converges to some unique optimal stationary point in the sense of ideal convergence if the ideal is invariant under translations. This kind of convergence generalizes, for example, statistical convergence and convergence with respect to logarithmic density zero sets.

Depending on magnitude and duration, any manoeuvering overload can damage the structure of an aircraft and adversely affect the pilot’s concentration and reaction time. These are all threats to flight safety. The flight safety envelope estimation method based on the classical reachable set cannot take into account the effect of manoeuvering overload. To overcome this limitation, a generalized reachable set known as a cost-limited reachable set is introduced into the computation of flight safety envelopes in this paper. It differs from the classical reachable set in that the performance index of the system can be set as the time integral of a running cost, and it can discuss the ability to reach the trim set before the performance index grows to the admissible cost. When computing the flight safety envelope, the running cost is set as a weighted sum of time consumption and manoeuver overload factor, and the flight safety envelope is defined as a cost-limited reachable set of the trim set. The flight safety envelopes and optimal control laws under the different weight of manoeuver overload factors are analyzed.

In this paper, a method based on neural networks for intelligently extracting weighting matrices of the optimal controllers’ cost function is presented. Despite the optimal and robust performance of controllers with the cost function, adjusting their gains which are the weighting matrices for the system state variables vector and the system inputs vector, is a challenging and time-consuming task that is usually selected by trial and error method for each specific application; and even little changes in the weighting matrices significantly impact problem-solving and system optimization. Therefore, it is necessary to select these gains automatically to improve controller performance and delete human energy to find the best gains. As a linear controller, linear quadratic regulator, and as a nonlinear controller, nonlinear model predictive control have been employed with trained networks to track the path of a wheeled mobile robot. The simulation and experimental results have been extracted and compared to validate the proposed method. These results have been demonstrated that the intelligent controller’s operation has lower error than the conventional method, which works up to 7% optimal in tracking and up to 19% better in angle state error; furthermore, as the most important aim, the required time and effort to find the weighting matrices in various situations has been omitted.

In this paper, we consider the problem of sustainable harvesting. We explain how the manager maximizes his/her profit according to the quantity of natural resource available in a harvesting area and under the constraint of penalties and fines when the quota is exceeded. We characterize the optimal values and some optimal strategies using a verification result. We then show by numerical examples that this optimal strategy is better than naive ones. Moreover, we define a level of fines which insures the double objective of the sustainable harvesting: a remaining quantity of available natural resource to insure its sustainability and an acceptable income for the manager.

The stabilisation and control mechanisms of an Unmanned Aerial System (UAS) must be properly designed to ensure acceptable flight performance. During their operation, these mechanisms are subjected to unknown and random environmental effects, making it imperative that all available information should be taken into consideration during the mechanisms’ design process (e.g. system dynamics, actuators, flight conditions and certain criteria requirements such as phugoid and short modes for longitudinal dynamics, and roll subsidence, spiral and Dutch-roll modes for lateral dynamics) in order to guarantee flight stability. Therefore, this paper introduces a novel methodology for the stabilisation and control of the UAS-S45 Bálaam, designed and manufactured by Hydra Technologies. This methodology uses composite controllers that combine feedback Linear Quadratic Regulators (LQR) and Proportional Integral Feed-Forward (PI-FF) compensation controller for stabilisation and tracking tasks, respectively. Furthermore, a Generalised Extended State Observer was implemented to provide robustness to the closed loop dynamics by introducing disturbance compensation. Furthermore, an Adaptive Neuro-Fuzzy Inference System (ANFIS) was adopted to perform a gain scheduling by computing the gains of each composite controller for certain unknown trim conditions within a given flight domain. Finally, several numerical assessments were performed to highlight the efficiency of the proposed methodology.

The quaternion is a powerful and common tool to avoid singularity in rotational dynamics in three-dimensional (3D) space. Here it has been particularly used as an alternative to Euler angles and rotation matrix. The application of the quaternion is exercised in quadrotor modeling and control. It changes the dynamics and represents a singularity-free attitude model. Here for the first time (for the best knowledge of authors), the state-dependent differential Riccati equation (SDDRE) control has been implemented on the quaternion-based model of a quadcopter. The proposed control structure is capable of aerobatic flight, and the Pugachev’s Cobra maneuver is chosen to assess the capability of the quaternion-based SDDRE approach. The introduced control simulator is validated by comparison with conventional dynamics based on Euler angles, controlled using a proportional-derivative (PD) controller on a normal regulation flight. The simulator successfully performed the Cobra maneuver and also validated the proposed structure. The more precision in regulation along with lower energy consumption demonstrated the superiority of the introduced approach.

This paper presents an integrated optimal control framework for velocity and steering control of an autonomous pursuit vehicle, where the control objectives satisfy the requirements of collision avoidance and moving target tracking. A distinctive feature of the proposed velocity and steering control is the application of logarithmic penalty functions to both. The control barrier imposed by logarithmic function provides a unique tool in computing a balanced trajectory with optimal tracking error, control effort and safety margin. Trajectories compliant with the safety regulations for autonomous driving have been planned based on estimated intention of the target and the obstacles. Effects of the controller weights have been extensively simulated to assess the performance of the proposed strategy in a variety of dynamic situations. The controller has been validated on a real-life robot by using a shrinking horizon control policy for iterative optimisation.

Although wearable robotic systems are designed to reduce the risk of low-back injury, it is unclear how effective assistance is, compared to improvements in lifting technique. We use a two-factor block study design to simulate how effective exoskeleton assistance and technical improvements are at reducing the risk of low-back injury when compared to a typical adult lifting a box. The effects of assistance are examined by simulating two different models: a model of just the human participant, and a model of the human participant wearing the SPEXOR exoskeleton. The effects of lifting technique are investigated by formulating two different types of optimal control problems: a least-squares problem which tracks the human participant’s lifting technique, and a minimization problem where the model is free to use a different movement. Different lifting techniques are considered using three different cost functions related to risk factors for low-back injury: cumulative low-back load (CLBL), peak low-back load (PLBL), and a combination of both CLBL and PLBL (HYB). The results of our simulations indicate that an exoskeleton alone can make modest reductions in both CLBL and PLBL. In contrast, technical improvements alone are effective at reducing CLBL, but not PLBL. The largest reductions in both CLBL and PLBL occur when both an exoskeleton and technical improvements are used. While all three of the lifting technique cost functions reduce both CLBL and PLBL, the HYB cost function offers the most balanced reduction in both CLBL and PLBL.

Microrobots with their promising applications are attracting a lot of attention currently. A microrobot with a triangular mechanism was previously proposed by scientists to overcome the motion limitations in a low-Reynolds number flow; however, the control of this swimmer for performing desired manoeuvres has not been studied yet. Here, we have proposed some strategies for controlling its position. Considering the constraints on arm lengths, we proposed an optimal controller based on quadratic programming. The simulation results demonstrate that the proposed optimal controller can steer the microrobot along the desired trajectory as well as minimize fluctuations of the actuators length.

Over the past few years, deep learning has risen to the foreground as a topic of massive interest, mainly as a result of successes obtained in solving large-scale image processing tasks. There are multiple challenging mathematical problems involved in applying deep learning: most deep learning methods require the solution of hard optimisation problems, and a good understanding of the trade-off between computational effort, amount of data and model complexity is required to successfully design a deep learning approach for a given problem.. A large amount of progress made in deep learning has been based on heuristic explorations, but there is a growing effort to mathematically understand the structure in existing deep learning methods and to systematically design new deep learning methods to preserve certain types of structure in deep learning. In this article, we review a number of these directions: some deep neural networks can be understood as discretisations of dynamical systems, neural networks can be designed to have desirable properties such as invertibility or group equivariance and new algorithmic frameworks based on conformal Hamiltonian systems and Riemannian manifolds to solve the optimisation problems have been proposed. We conclude our review of each of these topics by discussing some open problems that we consider to be interesting directions for future research.

A multi-agent engagement scenario is considered in which a high-value aircraft launches two defenders to intercept two homing missiles aimed at the aircraft. Under the assumption that all aircrafts have first-order linear dynamic characteristics, a combined multiple-mode adaptive estimation (MMAE) and a two-way cooperative optimal guidance law are proposed for the target–defenders team. Considering the full cooperation of the target and both the two defenders, the two-way cooperative strategies provide the analytical expressions for their optimal control input, enabling the target–defenders team to intercept the missiles with minimal control effort. To successfully intercept the missiles, MMAE is used to identify the guidance laws adopted by the missiles and estimate their states. The simulation results show that the target cooperating with the defenders to perform lure manoeuvres for the missiles can improve the guidance performance of the defenders as well as reduce the control effort of the defenders for intercepting the missiles.

In this paper, a new approach is presented for perfect torque compensation of the robot in point-to-point motions. The proposed method is formulated as an open-loop optimal control problem. The problem is defined as optimal trajectory planning with adjustable design parameters to compensate applied torques of a planar 5R parallel robot for a given task, perfectly. To illustrate the effectiveness of the approach, the obtained optimal path is used as the reference command in the experiment. The experimental outputs show that the performance index has been reduced by over 80% compared to the typical design of the robot.