arXiv subjects

Rushikesh Kamalapurkar

Publications and source records attributed to Rushikesh Kamalapurkar.

At least 37 records · Page 2Linked to original sources

arXiv2025

Lyapunov-Based Physics-Informed Deep Neural Networks with Skew Symmetry Considerations

Deep neural networks (DNNs) are powerful black-box function approximators which have been shown to yield improved performance compared to traditional neural network (NN) architectures. However, black-box algorithms do not incorporate known physics of the system and can yield results which are physically implausible. Physics-informed neural networks (PINNs) have grown in popularity due to their ability to leverage known physical principles in the learning process which has been empirically shown to improve performance compared to traditional black-box methods. This paper introduces the first physics-informed DNN controller for an Euler-Lagrange dynamic system where the adaptation laws are designed using a Lyapunov-based stability analysis to account for the skew-symmetry property of the inertia matrix and centripetal-Coriolis matrix. A Lyapunov-based stability analysis is provided to guarantee asymptotic convergence of the tracking error and the skew-symmetric prediction error. Simulations indicate that the developed update law demonstrates improvement in individual and overall function approximation capabilities when compared to a physics-informed adaptation law which does not incorporate knowledge of system symmetries.

eess.SY↗

arXiv2025

Safe Output-Feedback Adaptive Optimal Control of Affine Nonlinear Systems

In this paper, we develop a safe control synthesis method that integrates state estimation and parameter estimation within an adaptive optimal control (AOC) and control barrier function (CBF)-based control architecture. The developed approach decouples safety objectives from the learning objectives using a CBF-based guarding controller where the CBFs are robustified to account for the lack of full-state measurements. The coupling of this guarding controller with the AOC-based stabilizing control guarantees safety and regulation despite the lack of full state measurement. The paper leverages recent advancements in deep neural network-based adaptive observers to ensure safety in the presence of state estimation errors. Safety and convergence guarantees are provided using a Lyapunov-based analysis, and the effectiveness of the developed controller is demonstrated through simulation under mild excitation conditions.

eess.SY↗

arXiv2025

A Taylor Series Approach to Correction of Input Errors in Gaussian Process Regression

Gaussian Processes (GPs) are widely recognized as powerful non-parametric models for regression and classification. Traditional GP frameworks predominantly operate under the assumption that the inputs are either accurately known or subject to zero-mean noise. However, several real-world applications such as mobile sensors have imperfect localization, leading to inputs with biased errors. These biases can typically be estimated through measurements collected over time using, for example, Kalman filters. To avoid recomputation of the entire GP model when better estimates of the inputs used in the training data become available, we introduce a technique for updating a trained GP model to incorporate updated estimates of the inputs. By leveraging the differentiability of the mean and covariance functions derived from the squared exponential kernel, a second-order correction algorithm is developed to update the trained GP models. Precomputed Jacobians and Hessians of kernels enable real-time refinement of the mean and covariance predictions. The efficacy of the developed approach is demonstrated using two simulation studies, with error analyses revealing improvements in both predictive accuracy and uncertainty quantification.

eess.SY↗

arXiv2025

Improved Dwell-times for Switched Nonlinear Systems using Memory Regression Extension

This paper presents a switched systems approach for extending the dwell-time of an autonomous agent during GPS-denied operation by leveraging memory regressor extension (MRE) techniques. To maintain accurate trajectory tracking despite unknown dynamics and environmental disturbances, the agent periodically acquires access to GPS, allowing it to correct accumulated state estimation errors. The motivation for this work arises from the limitations of existing switched system approaches, where increasing estimation errors during GPS-denied intervals and overly conservative dwell-time conditions restrict the operational efficiency of the agent. By leveraging MRE techniques during GPS-available intervals, the developed method refines the estimates of unknown system parameters, thereby enabling longer and more reliable operation in GPS-denied environments. A Lyapunov-based switched-system stability analysis establishes that improved parameter estimates obtained through concurrent learning allow extended operation in GPS-denied intervals without compromising closed-loop system stability. Simulation results validate the theoretical findings, demonstrating dwell-time extensions and enhanced trajectory tracking performance.

eess.SY↗

arXiv2025

On Dynamic Mode Decomposition of Control-affine Systems

This paper builds on the theoretical foundations for dynamic mode decomposition (DMD) of control-affine dynamical systems by leveraging the theory of vector-valued reproducing kernel Hilbert spaces (RKHSs). Specifically, control Liouville operators and control occupation kernels are used to separate the drift dynamics from the input dynamics. A provably convergent finite-rank estimation of a compact control Liouville operator is obtained, provided sufficiently rich data are available. A matrix representation of the finite-rank operator is used to construct a data-driven representation of its singular values, left singular functions, and right singular functions. The singular value decomposition is used to generate a data-driven model of the control-affine nonlinear system. The developed method generates a model that can be used to predict the trajectories of the system in response to any admissible control input. Numerical experiments are included to demonstrate the efficacy of the developed technique.

eess.SY↗

arXiv2025

Gaussian Process-Based Scalar Field Estimation in GPS-Denied Environments

This paper presents a methodology for an autonomous agent to map an unknown scalar field in GPS-denied regions. To reduce localization errors, the agent alternates between GPS-enabled and GPS-denied areas while collecting measurements. User-defined error bounds determine the dwell time in each region. A switching trajectory is then designed to ensure field measurements in GPS-denied regions remain within the specified error limits. A Lyapunov-based stability analysis guarantees bounded error trajectories while tracking the desired path. The effectiveness of the proposed methodology is demonstrated through simulations, with an error analysis comparing the GP-predicted scalar field model to the actual field.

eess.SY↗

arXiv2025

Switched Optimal Control with Dwell Time Constraints

This paper presents an embedding-based approach for solving switched optimal control problems (SOCPs) with dwell time constraints. At first, an embedded optimal control problem (EOCP) is defined by replacing the discrete switching signal with a continuous embedded variable that can take intermediate values between the discrete modes. While embedding enables solutions of SOCPs via conventional techniques, optimal solutions of EOCPs often involve nonexistent modes and thus may not be feasible for the SOCP. In the modified EOCP (MEOCP), a concave function is added to the cost function to enforce a bang-bang solution in the embedded variable, which results in feasible solutions for the SOCP. However, the MEOCP cannot guarantee the satisfaction of dwell-time constraints. In this paper, a MEOCP is combined with a filter layer to remove switching times that violate the dwell time constraint. Insertion gradients are used to minimize the effect of the filter on the optimal cost.

math.OC↗

arXiv2024

Online Simultaneous State and Parameter Estimation for Second-order Nonlinear Systems

In this paper, a concurrent learning based adaptive observer is developed for a class of second-order nonlinear time-invariant systems with uncertain dynamics. The developed technique results in simultaneous online state and parameter estimation. A Lyapunov-based analysis is used to show that the state and parameter estimation errors are uniformly ultimately bounded. As opposed to persistent excitation which is required for parameter estimation in traditional adaptive control methods, the developed technique only requires excitation over a finite time interval.

eess.SY↗

arXiv2024

On Convergent Dynamic Mode Decomposition and its Equivalence with Occupation Kernel Regression

This paper presents a new technique for norm-convergent dynamic mode decomposition of deterministic systems. The developed method utilizes recent results on singular dynamic mode decomposition where it is shown that by appropriate selection of domain and range Hilbert spaces, the Liouville operator (also known as the Koopman generator) can be made to be compact. In this paper, it is shown that by selecting appropriate collections of finite basis functions in the domain and the range, a novel finite-rank representation of the Liouville operator may be obtained. It is also shown that the model resulting from dynamic mode decomposition of the finite-rank representation is closely related to regularized regression using the so-called occupation kernels as basis functions.

eess.SY↗

arXiv2024

Scalar Field Mapping with Adaptive High-Intensity Region Avoidance

This research is motivated by a scenario where a group of UAVs is assigned to map an unknown scalar field, with the imperative of maintaining a safe distance from the sources of the field to evade detection or damage. The location of the sources is unknown a priori, so the UAVs rely on measurements of the field intensity to gauge safety. The UAVs estimate the unknown scalar field using Gaussian process (GP) regression and use the estimate to generate a map of high-intensity regions using Hough transform (HT), updated online based on the field measurements. A convergence analysis shows the boundedness of the error between the actual scalar field and the learned scalar field. The effectiveness of the method is evaluated through simulations, showcasing its ability to accurately learn scalar fields with multiple high-intensity regions while reducing the number of measurements taken inside the high-intensity regions.

eess.SY↗

arXiv2024

An adaptive optimal control approach to monocular depth observability maximization

This paper presents an integral concurrent learning (ICL)-based observer for a monocular camera to accurately estimate the Euclidean distance to features on a stationary object, under the restriction that state information is unavailable. Using distance estimates, an infinite horizon optimal regulation problem is solved, which aims to regulate the camera to a goal location while maximizing feature observability. Lyapunov-based stability analysis is used to guarantee exponential convergence of depth estimates and input-to-state stability of the goal location relative to the camera. The effectiveness of the proposed approach is verified in simulation, and a table illustrating improved observability is provided.

eess.SY↗

arXiv2024

Nonuniqueness and Convergence to Equivalent Solutions in Observer-based Inverse Reinforcement Learning

A key challenge in solving the deterministic inverse reinforcement learning (IRL) problem online and in real-time is the existence of multiple solutions. Nonuniqueness necessitates the study of the notion of equivalent solutions, i.e., solutions that result in a different cost functional but same feedback matrix, and convergence to such solutions. While offline algorithms that result in convergence to equivalent solutions have been developed in the literature, online, real-time techniques that address nonuniqueness are not available. In this paper, a regularized history stack observer that converges to approximately equivalent solutions of the IRL problem is developed. Novel data-richness conditions are developed to facilitate the analysis and simulation results are provided to demonstrate the effectiveness of the developed technique.

eess.SY↗

arXiv2024

Dynamic Mode Decomposition with Control Liouville Operators

This paper builds the theoretical foundations for dynamic mode decomposition (DMD) of control-affine dynamical systems by leveraging the theory of vector-valued reproducing kernel Hilbert spaces (RKHSs). Specifically, control Liouville operators and control occupation kernels are introduced to separate the drift dynamics from the input dynamics. A given feedback controller is represented through a multiplication operator and a composition of the control Liouville operator and the multiplication operator is used to express the nonlinear closed-loop system as a linear total derivative operator on RKHSs. A spectral decomposition of a finite-rank representation of the total derivative operator yields a DMD of the closed-loop system. The DMD generates a model that can be used to predict the trajectories of the closed-loop system. For a large class of systems, the total derivative operator is shown to be compact provided the domain and the range RKHSs are selected appropriately. The sequence of models, resulting from increasing-rank finite-rank representations of the compact total derivative operator, are shown to converge to the true system dynamics, provided sufficiently rich data are available. Numerical experiments are included to demonstrate the efficacy of the developed technique.

math.OC↗

arXiv2023

Pilot Performance modeling via observer-based inverse reinforcement learning

The focus of this paper is behavior modeling for pilots of unmanned aerial vehicles. The pilot is assumed to make decisions that optimize an unknown cost functional, which is estimated from observed trajectories using a novel inverse reinforcement learning (IRL) framework. The resulting IRL problem often admits multiple solutions. In this paper, a recently developed novel IRL observer is adapted to the pilot modeling problem. The observer is shown to converge to one of the equivalent solutions of the IRL problem. The developed technique is implemented on a quadcopter where the pilot is a linear quadratic supervisory controller that generates velocity commands for the quadcopter to travel to and hover over a desired location. Experimental results demonstrate the robustness of the method and its ability to learn equivalent cost functionals.

eess.SY↗

arXiv2023

Online Observer-Based Inverse Reinforcement Learning

In this paper, a novel approach to the output-feedback inverse reinforcement learning (IRL) problem is developed by casting the IRL problem, for linear systems with quadratic cost functions, as a state estimation problem. Two observer-based techniques for IRL are developed, including a novel observer method that re-uses previous state estimates via history stacks. Theoretical guarantees for convergence and robustness are established under appropriate excitation conditions. Simulations demonstrate the performance of the developed observers and filters under noisy and noise-free measurements.

eess.SY↗

arXiv2023

Fault Detection via Occupation Kernel Principal Component Analysis

The reliable operation of automatic systems is heavily dependent on the ability to detect faults in the underlying dynamical system. While traditional model-based methods have been widely used for fault detection, data-driven approaches have garnered increasing attention due to their ease of deployment and minimal need for expert knowledge. In this paper, we present a novel principal component analysis (PCA) method that uses occupation kernels. Occupation kernels result in feature maps that are tailored to the measured data, have inherent noise-robustness due to the use of integration, and can utilize irregularly sampled system trajectories of variable lengths for PCA. The occupation kernel PCA method is used to develop a reconstruction error approach to fault detection and its efficacy is validated using numerical simulations.

stat.ML↗

arXiv2023

State and Parameter Estimation for Affine Nonlinear Systems

Real-world control applications in complex and uncertain environments require adaptability to handle model uncertainties and robustness against disturbances. This paper presents an online, output-feedback, critic-only, model-based reinforcement learning architecture that simultaneously learns and implements an optimal controller while maintaining stability during the learning phase. Using multiplier matrices, a convenient way to search for observer gains is designed along with a controller that learns from simulated experience to ensure stability and convergence of trajectories of the closed-loop system to a neighborhood of the origin. Local uniform ultimate boundedness of the trajectories is established using a Lyapunov-based analysis and demonstrated through simulation results, under mild excitation conditions.

eess.SY↗

arXiv2023

The kernel perspective on dynamic mode decomposition

This manuscript revisits theoretical assumptions concerning dynamic mode decomposition (DMD) of Koopman operators, including the existence of lattices of eigenfunctions, common eigenfunctions between Koopman operators, and boundedness and compactness of Koopman operators. Counterexamples that illustrate restrictiveness of the assumptions are provided for each of the assumptions. In particular, this manuscript proves that the native reproducing kernel Hilbert space (RKHS) of the Gaussian RBF kernel function only supports bounded Koopman operators if the dynamics are affine. In addition, a new framework for DMD, that requires only densely defined Koopman operators over RKHSs is introduced, and its effectiveness is demonstrated through numerical examples.

math.FA↗