Search

Gradient-based optimization and control frameworks have been utilized in several applications. The learning rate parameter is typically chosen following a schedule or using methods such as line search to enhance the convergence rate. Recently, the machine learning community has developed methodologies for automated tuning of the learning rate, known as adaptive gradient methods. This paper develops a control theory-inspired framework for modeling adaptive gradient methods that solve non-convex optimization problems. We first model the adaptive gradient methods in a state–space framework, which allows us to present simpler convergence proofs of prominent adaptive optimizers, such as AdaGrad, Adam, and AdaBelief. The proposed framework is constructive because it allows synthesizing new adaptive optimizers. To illustrate this fact, we then utilize the transfer function paradigm from classical control to propose a new variant of Adam, coined AdamSSM, and prove its convergence. We add an appropriate pole-zero pair in the transfer function from squared gradients to the second moment estimate. Applications on benchmark machine learning tasks of image classification using CNN architectures and language modeling using LSTM architecture demonstrate that the AdamSSM algorithm improves the gap between generalization accuracy and faster convergence than the recent adaptive gradient methods. [ABSTRACT FROM AUTHOR]

Design of a physical system and its controller has significant ramifications on the overall system performance. The traditional approach of first optimizing the physical design and then the controller may lead to sub-optimal solutions. This is due to the interdependence between the physical design and control parameters through the dynamic equations. Recognition of this fact paved the way for investigation into the “Co-Design” research theme wherein the overall system’s physical design and control are simultaneously optimized. In this paper, a novel approach to address the co-design problem for a class of Linear Time Invariant (LTI) dynamic systems controlled by a Linear Quadratic Regulator (LQR) feedback is presented. The considered co-design problem is formulated as a non-convex optimization problem with Algebraic Riccati Equation (ARE) constraint and convex design objective function. Using Semi-Definite Programming (SDP) duality, the ARE constraint is reduced into equivalent Bilinear Matrix Inequality (BMI) constraints. This reformulated co-design problem is solved using an iterative algorithm based on the Generalized Benders Decomposition (GBD) and Gradient Projection Method. The proposed algorithm converges to a solution which is within a specified tolerance from the nearest local minimum (in special cases global minimum) in a finite number of iterations. Necessary and sufficient conditions are developed to test minimality. Three examples are presented to show efficacy of the proposed algorithm. [ABSTRACT FROM AUTHOR]

Abstract: Networked control of robotic systems is widely recognized as a potentially transformative technological enabler for several applications. However, the issues of time delays in communication and recovery from data losses have emerged as the pivotal issues that have stymied practical deployment. The study for control of robotic system with input/output communication delays has attracted many researchers’ attention, but the existing results have been primarily developed for rigid-joint robots. Since joint flexibility is largely unavoidable in practical manipulators, in this paper the set-point control problem for flexible-joint robots with input/output communication delays is studied. It is demonstrated that the scattering variables address the stability problem for unknown constant delays, however, in contrast to the rigid-robot case, they cannot guarantee set-point regulation. In addition, we compute the explicit dependence of the regulation errors on the communication delays, control gains, and the desired set-point configuration. Without exact knowledge of time delays, a scattering variable based controller with position feedback is subsequently studied in this paper to guarantee stability with improved regulation performance. The control architecture is further extended to the case with time-varying delays. Simulation results are presented to validate the efficacy of the proposed control algorithms. [Copyright &y& Elsevier]

Abstract: In teleoperation systems operating in complex environments, due to the cognitive limitations of the human operator and lack of complete information about the remote environment, safety and performance of such systems can potentially be comprised. In order to ensure the safety and enhance the efficiency of complex teleoperation systems operating in cluttered environments, in this paper we investigate a semi-autonomous control framework for bilateral teleoperation. The semi-autonomous teleoperation system is composed of heterogeneous master and slave robots, where the slave robot is assumed to be a redundant manipulator. Considering robots with different configurations, and in the presence of dynamic uncertainties and asymmetric communication delays, we first develop a control algorithm to ensure position and velocity tracking in the task space. Additionally in the absence of dynamic uncertainty, and in the presence of human operator and environmental forces, all signals of the proposed teleoperation system are proven to be ultimately bounded. The redundancy of the slave robot is then utilized for achieving autonomous sub-task control, such as singularity avoidance, joint limits, and collision avoidance. The control algorithms for the proposed semi-autonomous teleoperation system are validated through numerical simulations on a non-redundant master and a redundant slave robot. [Copyright &y& Elsevier]

Abstract: Bilateral teleoperators, designed within the passivity framework using concepts of scattering and two-port network theory, provide robust stability against constant delay in the network and velocity tracking, but cannot guarantee position tracking in general. In this paper we fundamentally extend the passivity-based architecture to guarantee state synchronization of master/slave robots in free motion independent of the constant delay and without using the scattering transformation. We propose a novel adaptive coordination architecture which uses state feedback to define a new passive output for the master and slave robots containing both position and velocity information. A passive coordination control is then developed which uses the new outputs to state synchronize the master and slave robots in free motion. The proposed algorithm also guarantees ultimate boundedness of the master/slave trajectories on contact with a passive environment. Experimental results are also presented to verify the efficacy of the proposed algorithms. [Copyright &y& Elsevier]

This paper considers the multi-agent linear least-squares problem in a server–agent network architecture. The system comprises multiple agents, each with a set of local data points. The agents are connected to a server, and there is no inter-agent communication. The agents' goal is to compute a linear model that optimally fits the collective data. The agents, however, cannot share their data points. In principle, the agents can solve this problem by collaborating with the server using the server–agent network variant of the classical gradient-descent method. However, when the data points are ill-conditioned , the gradient-descent method requires a large number of iterations to converge. We propose an iterative pre-conditioning technique to mitigate the deleterious impact of the data points' conditioning on the convergence rate of the gradient-descent method. Unlike the conventional pre-conditioning techniques, the pre-conditioner matrix used in our proposed technique evolves iteratively. We show that our proposed algorithm converges linearly with an improved rate of convergence in comparison to both the classical and the accelerated gradient-descent methods. For the special case, when the solution of the least-squares problem is unique, our algorithm converges to the solution superlinearly. Through numerical experiments on benchmark least-squares problems, we validate our theoretical findings, and also demonstrate our algorithm's improved robustness against process noise. [ABSTRACT FROM AUTHOR]

Introduction: Vasculobiliary and vascular injuries following cholecystectomy are the most serious complications requiring complex surgical management resulting in greater patient morbidity and mortality.Methodology: The study was performed at a tertiary teaching hospital of North India. Records of patients referred for biliary or vascular injury sustained during cholecystectomy were reviewed retrospectively to identify patients with vascular injury between January 2009 and March 2018. Clinical profile, hospital course and outcome of these patients were analysed.Results: Over nine years, 117 patients were referred for cholecystectomy related complications. Total incidence of vascular injury was 5.1% (6/117). Combined vasculobiliary injury (VBI) occurred in 3.4% (4/117) while isolated vascular injury was present in 1.7% patients (2/117). Most (5/6) patients were operated for uncomplicated gall stone disease. Incidences of portal vein (PV) and right hepatic artery (RHA) injuries were equal (3/6). PV injuries were repaired either during cholecystectomy (1/3) or during re-exploration after damage control packing (2/3). RHA injuries presented as pseudoaneurysm and were managed surgically (2/3) or by coil embolization (1/3). All VBI referrals (4/117) were following open cholecystectomy. In VBI patients, vascular injury was diagnosed intra-operatively in two while it was diagnosed several weeks after cholecystectomy in two others. Biliary injury manifested as bile leak post-operatively in all four of them. Nature of biliary injury could be characterized in only 50% (2/4) patients. Definitive repair of biliary injury was performed in one patient only. There was one mortality in our series.Conclusion: Vascular injury is an uncommon complication of cholecystectomy with catastrophic outcome if not managed timely and properly. Adequate surgeon training, keeping the possibility of aberrant vasculobiliary anatomy in all cases, and proper surgical technique is crucial for prevention of such injuries. However in such an event, proper documentation and referral to tertiary centre will help in decreasing morbidity and further litigation. [ABSTRACT FROM AUTHOR]