Titles and Abstracts

Title:  Near-optimal solutions of convex semi-infinite programs via targeted sampling 

Speaker:   Ashwin Aravind (IIT Bombay)

Abstract: Semi-infinite programs (SIP) are encountered in a diverse spectrum of applications ranging from robotics and control to finance. We propose an approach to find the optimal value of a convex SIP that involves identifying a finite set of relevant constraints by solving a finite-dimensional global maximization problem. One of the major advantages of our approach is that it admits a plug-and-play module where any suitable global optimization algorithm can be employed to obtain the optimal value of the SIP. As an example, we propose a simulated-annealing based algorithm which is useful especially when the constraint index set is high-dimensional.  

Title: Route planning for capacity-restricted agents over a railway network without disrupting train schedules 

Speaker:  Somnath Buriuly (IIT Bombay)

Abstract: Deploying mobile instrumentation for railway track inspection without disrupting regular train schedules is a routing and scheduling problem. We formulate this problem in a discrete-optimization framework and propose variants of the column generation matheuristic. Our findings validate the matheuristics and compare their bound improvements with respect to problem size. 

Title: Adaptive identification of linear infinite-dimensional systems 

Speaker:  Sudipta Chattopadhyay (IIT Bombay)

Abstract: Linear infinite-dimensional systems are widely used to model physical plants and identifying the unknown parameter in these models is a problem of fundamental interest in many areas of engineering. There are two distinct approaches to this problem: non-adaptive and adaptive. The adaptive approach offers certain advantages: the adaptive algorithms can be used to track slowly time-varying parameters and can potentially be used in the design of adaptive controllers. In this talk, we will discuss a novel adaptive parameter identification algorithm which is applicable to a large class of linear single-input single-output (SISO) infinite-dimensional systems. 

Title: A new framework for detecting actuator attacks 

Speaker:  Souvik Das (IIT Bombay)

Abstract: Cyber-Physical systems (CPSs) monitor and regulate several critical large-scale infrastructures such as smart grids, transportation systems, and wearable medical systems. Recent instances of cyber-attacks, such as the Stuxnet computer worm attack in Iran and the cyber assault on Ukraine's power grid demonstrate the security vulnerabilities of large-scale CPSs. With increasing complexities of the CPSs around us, the possibilities available to the attackers to launch sophisticated and intelligent attacks have increased.  Consequently, there is an emergent need to devote considerable attention to the issue of security of CPSs. This talk aims to address the problem of detecting the presence of intelligent and adversarial agents in CPSs; particularly focusing on arbitrarily clever actuator attacks.  

Title: Information Revelation Through Signalling

Speaker:  Reema Deori (IIT Bombay)

Abstract: We study a Stackelberg game between a sender (leader) and a receiver (follower) where the sender (the informed player) attempts to shape the information of a less informed receiver (follower) who in turn takes an action that determines the payoff for both players. We show that the concept of  list coloring property of a graph can be used to characterize the  Stackelberg equilibrium strategies of the sender.  We introduce the concept of the sender graph and show that every list coloring policy of the sender graph is a Stackelberg equilibrium strategy of the sender and vice-versa. Our main contribution is the study of informativeness – the minimum amount of correct information that the receiver can obtain in any such equilibrium and we show that informativeness can be characterized by the list chromatic number of the sender graph.

Title: Robust model predictive control: explicit solutions for low- through moderate-dimensional linear systems 

Speaker:  Siddhartha Ganguly (IIT Bombay)

Abstract: Model predictive control (MPC) is one of the most applicable control techniques available today; it has witnessed explosive growth and proliferation in several industries. It is an optimization-based method designed for implicit feedback control of systems. MPC has witnessed intense activity over the past several decades, and both the theoretical and practical aspects of this technique are reasonably well-developed, especially in the linear regime. But computational tractability remains the primary bottleneck even in the linear regime, especially in the context of robust minmax MPC due to the need of solving a robust optimization problem repeatedly via shifting the prediction horizon.

Explicit MPC replaces the task of optimization at each time instant to that of a function evaluation --- the implicit feedback in MPC is given an explicit form by means of this technique. This talk is designed to expose entirely novel (and some yet unpublished) ideas, with emphasis on heuristic-free computationally tractable methods for explicit robust minmax MPC in the context of low- to moderate-dimensional linear systems leveraging several tools from approximation theory and deep neural networks.   

Title: Robustness of solutions of nonlinear equations: Mapping degree at work 

Speaker:  Ashutosh Jindal (IIT Bombay)

Abstract: In engineering application one often require to solve nonlinear equations. Parameteric distrubances may result in shifting the equillbrium point. Mapping degree allows us to show (non)existence of solutions of set of equations over a given domain. We utilize these reults to quantify the robustness of solution of nonlinear equation and provide estimates that require minimum computation efforts.  

Title:  Stabilization of cascade interconnection of an 1D heat equation in polar coordinates and an ODE using LQR design 

Speaker: S Bhargav Pavan Kumar (IIT  Bombay)

Abstract: We consider ODE-PDE cascade systems in which the input is applied to the ODE system whose output drives the PDE system and PDE-ODE cascade systems in which the input is applied to the PDE system whose output drives the ODE system. In both the cascade systems we take the PDE system to be an 1D heat equation in polar coordinates containing a singular term. We address the problem of designing state-feedback control laws for stabilizing these cascade systems via finite-dimensional approximation and LQR design. Using the Galerkin technique we first obtain an N th -order ODE approximation for the cascade system. We fix a quadratic cost function for the cascade system and consider the corresponding operator algebraic Riccati equation (ARE). We associate an appropriate matrix ARE with the ODE approximation. Under some natural assumptions on the cascade system, we verify that certain well-known hypothesis in the literature hold, which imply that the solution Π N of the matrix ARE converges to the solution Π of the operator ARE strongly as N tends to infinity. 

Title: Data-driven Robust Optimization for Energy-Aware Trajectory Planning of Electric Vehicles 

Speaker:  Simran Kumari (IIT Kharagpur)

Abstract: Our talk focuses on energy aware safe trajectory planning functionality of Advanced Driver Assistance System (ADAS) for electric vehicles. ADAS is an essential technology for electric vehicles in order to be able to extend electric range in the face of battery storage limitation as well as for safe maneuver in an uncertain environment (traffic). We analyze the effect of lateral dynamics on battery energy consumption and consequently driving range. Motivated by the study, we present a Nonlinear Model Predictive Control formulation utilizing a data-driven robust framework to tackle energy optimal and safe navigation of electric vehicles. 

Title: Epidemic Propagation under Game-theoretic Activation, Testing and Vaccination 

Speaker:  Urmee Maitra (IIT Kharagpur)

Abstract: We present a dynamic population game model to capture the behavior of a large population of individuals in presence of an infectious disease or epidemic. Individuals can be in one of five possible infection states at any given time: susceptible, asymptomatic, symptomatic, recovered and unknowingly recovered, and choose whether to opt for vaccination, testing or social activity with a certain degree. We define the evolution of the proportion of agents in each epidemic state, and the notion of best response for agents that maximize long run discounted expected reward as a function of the current state and policy. We further show the existence of a stationary Nash equilibrium and explore the transient evolution of the disease states and individual behavior under a class of evolutionary learning dynamics. Our results provide compelling insights into how individuals evaluate the tradeoff among vaccination, testing and social activity under different parameter regimes, and the impact of different intervention strategies (such as restrictions on social activity) on infection prevalence. 

Title: Optimal switching of a networked control system under communication constraints

Speaker: Harsh Oza (IIT Bombay)

Abstract: Networked Control Systems(NCS) comprise of a plant, a controller and a (wireless) communication network. Limitations on communication arises naturally, in form of bandwidth limitations, packet loss, delay etc. In this talk, the focus is on bandwidth limitation such that the plant and the controller share a channel, and communication is possible only in one way at a time. Our objective to find the optimal switching strategy is achieved using discrete time Pontryagin's Maximum Principle. 

Title: Cluster Consensus of Multi-agent Systems With Second Order Dynamics Over Matrix-weighted Graphs 

Speaker: Gopika R (BITS Pilani-Goa)

Abstract: We discusses the cluster consensus problem for a set of agents having second order dynamics connected by multi-partite matrix weighted interaction graph. Under appropriate connectedness conditions, an additional cluster consensus scheme is developed where agents update their states with respect to a dynamic common leader and form cluster consensus in position, tracking the position of the leader. In this case, the agent velocities are shown to converge to the leader velocity, asymptotically. Results are analytically shown using Lyapunov theory and are illustrated by numerical simulations. I 

Title: Strategic classification for non-uniform preferences 

Speaker: Manish Kumar Singh (IIT Bombay)

Abstract: We study the problem of strategic classification, where feature vectors are subject to adversarial manipulation by users who want to obtain favourable labels. We show that the naïve classifier, which is optimal in the absence of an adversary, is not robust to such manipulation. We consider a general setting of the problem where the adversary has a non-uniform preference for labels and can modify any feature vector at some cost. We formulate this as a screening game between the adversary and the classifier and derive the optimal classifier in this setting. 

Title: Collision Cone Control Barrier Function 

Speaker:  Manan Tayal  (IISc)

Abstract: The research focuses on a novel approach called Collision Cone Control Barrier Functions (C3BFs) to avoid collisions with dynamic obstacles in both ground and aerial vehicles. The proposed control barrier function formulation is inspired by collision cones used in trajectory planning to ensure safety by constraining the relative velocity between the vehicle and the obstacle to always point away from each other. The efficacy of this approach is demonstrated through simulations and hardware implementations on Copernicus, and Crazyflie 2.1, displaying its effectiveness in avoiding collisions with dynamic obstacles. Real-time implementation is achieved using Quadratic Programs (QPs) known as CBF-QPs. Comparative analysis with existing CBF-QPs, particularly higher-order CBF-QPs (HO-CBF-QPs), highlights the less conservative nature of the proposed approach. Overall, this researching unmanned vehicles, enhancing their manoeuvrability and safety in dynamic environments.