| |
| WeBT1 |
Auditorium |
| Control Applications II |
Regular Session |
| |
| 10:30-10:50, Paper WeBT1.1 | |
>Adaptive Fuzzy Predictor Based Fast Terminal Sliding Mode Controller Design for Two-Link Robot Manipulator |
|
| Kumar, Shreyas | Indian Institute of Engineering Science and Technology, Shibpur |
| Dash, Bishesha | Indian Institute of Engineering Science and Technology, Shibpur |
| Maiti, Roshni | Indian Institute of Engineering Science and Technology, Shibpur |
| Guha, Dipayan | Motilal Nehru National Institute of Technology Allahabad, U.P., |
Keywords: Stability of nonlinear systems, Predictive control for nonlinear systems, Mechanical systems/robotics
Abstract: This paper proposes an Adaptive Fuzzy Predictor based Fast Terminal Sliding Mode Control (AFP-FTSMC) for a two-link robotic manipulator (TLRM). The unknown disturbances, changing with time, present in the TLRM system are estimated using the fuzzy predictor model and the accurate control action is formulated using the FTSM controller to tackle the nonlinearities in the system. The disturbance, varying with time, is estimated using adaptive fuzzy predictor, and the fuzzy adaptive rule is developed from the Lyapunov stability criteria. The chattering phenomenon and singularity problem is efficiently addressed by utilizing the fast terminal sliding mode control law. The overall controller is capable of producing fast-tracking performance by rejecting the disturbance and chattering adequately. The stable performance of the TLRM system with the proposed controller is guaranteed by utilizing Lyapunov stability criteria. The experimental results demonstrate that the proposed controller achieves superior tracking performance and robustness for nonlinear TLRM system compared to the existing terminal sliding mode control method, even in the presence of significant uncertainties and time-varying external disturbances.
|
| |
| 10:50-11:10, Paper WeBT1.2 | |
>Stability Analysis Using Describing Function for Non-Linear Attitude Control Loop with Slosh Dynamics of Chandrayaan-3 Landing Mission |
|
| Rallapalli, Aditya | UR Rao Satellite Center |
| Chakrabarti, Debajyoti | Indian Space Research Organisation (ISRO) |
| Priyadarshini, Nivriti | UR Rao Satellite Centre |
| Mp, Rijesh | UR Rao Satellite Centre |
| Gvp, Bharat Kumar | UR Rao Satellite Centre |
Keywords: Aerospace, Control applications, Spacecraft control
Abstract: Chandrayaan-3 became the first Indian Lunar Lander mission which landed on Moon’s south pole in 2023. The lunar landing mission consists of multiple phases where primary deceleration was done using liquid engines. The navigation, guidance, and control system dictates the trajectory and descent operations. Guidance generates the attitude reference profile to be tracked by the inner control loop. Attitude control is performed using non-linear reaction control system which operates in on-off mode. A high-fidelity mechanical analogous slosh (pendulum) model is used to analyze the sloshing phenomena and control stability during the powered descent phase. Propellant sloshing induces disturbance which can interact with the inner attitude control loop and impact the overall stability margins. The pulse width pulse frequency modulator (PWPFM) converts the linear control commands to non-linear on-off pulses for the reaction control system (RCS). In this paper, the linear stability analysis of the attitude control loop with propellant slosh dynamics is extended to incorporate the PWPFM non-linearity using the describing function analysis. An algorithm to numerically compute describing function is proposed which resolves the initialization issue with integrators. The stability margins and stability of limit cycle as a function of time has been computed. Finally, a correlation between frequency domain analyses and high-fidelity time domain simulation is presented.
|
| |
| 11:10-11:30, Paper WeBT1.3 | |
>Reinforcement Learning-Based Control of Multivariable Processes with Time Delays |
|
| Patil, Dinesh | ABB India Corporate Research Centre |
| Joshi, Tanuja | ABB |
| Pinnamaraju, Vivek Shankar | Asea Brown Boveri (ABB) |
| M, Sudhakar | ABB Global Industries & Services Pvt Ltd |
| Kubal, Nandkishor | ABB Global Industries and Services Limited |
Keywords: Agents-based systems, Control applications, Delay systems
Abstract: The dynamics of industrial processes are often characterized by significant and varied time delays. Direct application of conventional reinforcement learning architectures using feed-forward neural networks is inadequate, often resulting in deteriorated control performance. In this work, we embed a recurrent neural network (RNN) architecture in the reinforcement learning framework for the actor and critic networks to effectively handle multivariate process dynamics with large and varied delays. This approach enables the resulting trained control policy to understand that current actions affect the output(s) with certain delay(s). The efficacy of the proposed approach has been tested on simulation examples comprising single-input single-output (SISO) and multi-input multi-output (MIMO) systems with delays for set-point tracking and disturbance rejection scenarios. The results demonstrate the efficacy of the approach in effectively managing multivariate systems, even in the presence of inherent delays.
|
| |
| 11:30-11:50, Paper WeBT1.4 | |
>Motion Camouflage Guidance Strategies in an Aggressive Honey Bee |
|
| Jada, Chakravarthi | Rajiv Gandhi University of Knowledge Technologies |
| Urlana, Ashok | RGUKT Nuzvid |
| Baswani, Pavan | University |
| Shaik, Gouse Basha | Rajiv Gandhi University of Knowledge Technologies, Nuzvid |
| Rayavarapu, Sita Mounika | Rajiv Gandhi University of Knowledge Technologies |
| Peteru, Venkata Sai Rakesh | Rajiv Gandhi University of Knowledge and Technologies |
| Sandala, Santhi Sree | Rajiv Gandhi University of Knowledge Technologies |
Keywords: Biologically-inspired methods, Behavioral systems, Biological systems
Abstract: The paper presents the conduction of the experiments and motion-camouflaged approach to determine how a bee chases and intercepts a moving target. The experimental setup is made ready, and the bee-target motion episodes are recorded for various target movements. The bee-target position tuples are generated at all-time instances, followed by the simulation of trajectories. The verification of motion camouflage strategies by the bee is carried out. Then a kinematic model with three phases is hypothesized and compared with motioncamouflaged data and it is found that both are interrelated. Then, switching between camouflage strategies is concluded. Concluding remarks on the future scope were discussed at the end. This research work is the first of its kind in observing the aggressive behavior in the bees and looking into their motion strategies.
|
| |
| 11:50-12:10, Paper WeBT1.5 | |
>Discounted Evolutionary Stochastic Games and Replicator Dynamics |
|
| Murali, Divya | Indian Institute of Technology Bombay |
| Rao, K.S. Mallikarjuna | Indian Institute of Technology Bombay |
Keywords: Biological systems, Agents-based systems, Optimization
Abstract: In this paper, we study the discounted evolutionary stochastic game model. Using this model, we give an alternate definition for Nash equilibrium (NE) of stochastic games and also define evolutionarily stable strategy (ESS) for the same. Also, we provide a version of replicator dynamics for discounted stochastic games and prove that the defined ESS is asymptotically stable in it. Further, we explore finite stage stochastic game and study the relation between NE and replicator dynamics of the N-step stochastic game and the infinite horizon stochastic game.
|
| |
| 12:10-12:30, Paper WeBT1.6 | |
>Hybrid Digital Twin Based Model Predictive Control for Optimal Charging of Lithium-Ion Battery to Enhance SOH |
|
| V, Kaviya | Mit, Anna University |
| M A, Ahamed Atheeq | Mit, Anna University |
| S, Sutha | Madras Institute of Technology Campus, Anna University Chennai |
| Natarajan, Pappa | Professor, Department of Instrumentation Engineering, MIT Campu |
Keywords: Predictive control for nonlinear systems, Optimal control, Automotive
Abstract: Sophisticated control schemes play a crucial role in enhancing the State-of-Health (SOH) of Lithium-ion (Li-ion) batteries by generating an optimal charging profile with operating constraints such as charging time and battery lifespan. However, the development of model-based control strategies with constraints for highly critical systems like battery is challenging due to its complex behavior. In this paper, a hybrid digital twin-based Model Predictive Control (MPC) is proposed to generate an optimal charging profile for a 3.6V, 18Ah LiFePO4 battery. The hybrid digital twin is developed by combining physics-based Electro-Thermal Model (ETM) and linear regression model. The optimal control problem is formulated by imposing constraints on input current and batteries internal states and MPC is designed. The demonstrated simulation results show the performance of the proposed control scheme by satisfying the objective function of minimizing charging time and battery temperature to enhance SOH of battery.
|
| |
| WeBT2 |
Seminar Hall 1 |
| Robust and Adaptive Control II |
Regular Session |
| |
| 10:30-10:50, Paper WeBT2.1 | |
>Adaptive Control for Student Grading System: A Linear Control Approach |
|
| Jada, Chakravarthi | Rajiv Gandhi University of Knowledge Technologies |
| Guguloth, Bhargavi | Rajiv Gandhi University of Knowledge Technologies |
Keywords: Adaptive systems, Behavioral systems, Simulation
Abstract: Abstract—In the Realm of the Academic performance enhancement, study delves into the implementation of an Adaptive Control mechanism within a student Grading system. The background underscores the critical need for a system to optimize student Grades by dynamically adjusting study Efforts and improving teaching methodologies. The research aims to address the primary question of how Adaptive systems can positively influence student performance through behavioral and control education. The Methodology employs a Linear Control Approach utilizing Matlab simulations to model student academic performance and faculty guidance mechanisms. Results indicate that the adaptive control system effectively enhances student grades by tailoring educational experiences to individual needs and promoting continuous improvement. Ultimately, the study concludes that adaptive systems play a Pivotal role in guiding student development, fostering academic success, and facilitating a culture of educational excellence. Index Terms—Adaptive systems, Behavioural systems, Control education, Grade, Matlab Simulation, Feedback Control, A
|
| |
| 10:50-11:10, Paper WeBT2.2 | |
>Koopman Framework-Based Observer Design for Actuator Fault Identification in Quadrotor |
|
| Akumalla, Ravi Kiran | Indian Institute of Technology, Mandi |
| Shukla, Amit | Indian Institute of Technology Mandi |
| Jain, Tushar | Indian Institute of Technology Mandi |
Keywords: Modeling and simulation, Fault detection/accomodation, Observers for linear systems
Abstract: This paper investigates the implementation of the Koopman framework to quadrotor by numerical simulations. Initially, the nonlinear dynamics of a quadrotor is modelled in MATLAB/SIMULINK. The attitude and altitude control of the model is achieved using individual PID controllers. From this closed-loop model, the attitude data and angular torque data are collected. Using this data, a linear approximation of the attitude dynamics of the quadrotor is found by applying the Koopman framework. This identified Koopman linear model is leveraged as a predictor within the model predictive controller (MPC) to achieve inner loop attitude tracking of the quadrotor. Then, a loss of actuator effectiveness (LOE) fault is introduced in one of the actuators, which is successfully identified by designing a linear observer based on the identified Koopman linear model. Finally, the simulation results are provided, describing the attitude control and actuator fault identification in quadrotor using the Koopman framework.
|
| |
| 11:10-11:30, Paper WeBT2.3 | |
>Temperature Control of a PEMFC by Using Barrier Function Based Super-Twisting Sliding Mode Control |
|
| Meshram, Tushar | Visvesvaraya National Institute of Technology, Nagpur |
| Singh, Jay Prakash | VNIT Nagpur |
Keywords: Robust control, Process control, Control applications
Abstract: Hydrogen as a clean and sustainable energy carrier holds immense promise for addressing contemporary energy challenges. Proton Exchange Membrane Fuel Cells (PEMFC) represents a pivotal technology in the production of hydrogen through electrolysis of water. In this study, we explore the efficacy of two distinct control strategies-PI and Barrier function based super twisting Sliding Mode Control (BF-STSMC) in regulating the temperature of a PEMFC system. The control input which modulates the coolant flow rate is examined in response to variations in the current passing through the stack. Simulation results reveal significant insights into the performance of the mentioned control strategies. BF-STSMC approach demonstrates a shorter settling time and mitigates chattering highlighting its potential for efficient temperature regulation in PEMFC systems. The findings presented in this paper contribute to advancing our understanding of temperature control strategies in PEMFC systems and offer valuable insights for optimizing hydrogen production processes.
|
| |
| 11:30-11:50, Paper WeBT2.4 | |
>UDE Based Robust Predictive Control for CoM Tracking of Humanoid Robot |
|
| Kolhe, Jaywant | R&DE (E), Dighi, Pune, INDIA - 411015 |
| Uttam, Prakash K | Reasearch and Development Establishment (Engineers) |
| S, Soumya | Defence Institute of Advanced Technology |
| Talole, Sanjay | Defence Institute of Advanced Technology, Girinagar, Pune-411 02 |
Keywords: Robust control, Predictive control for linear systems, Reduced order modeling
Abstract: This work proposes a new robust walking control design for humanoid robot. Using a representative three dimensional linear inverted pendulum mode (3D-LIPM) of a humanoid, a Center of Mass (CoM) tracking controller is designed using Continuous Time Predictive Control (CTPC) approach. To achieve robustness, the controller is augmented by estimate of the disturbance obtained using Uncertainty and Disturbance Estimator (UDE). Closed loop stability of the overall system is established. Next, performance of the UDE based robust Continuous Time Predictive Control (UDE+CTPC) for CoM trajectory tracking in presence of external disturbances is demonstrated through simulations.
|
| |
| 11:50-12:10, Paper WeBT2.5 | |
>Almost Fixed-Time Consensus for Second-Order Multi-Agent System with Barrier Function Integral Sliding Mode Control |
|
| Swaraj, Tara | National Institute of Technology Silchar |
| Nath, Krishanu | National Institute of Technology Silchar |
| Bera, Manas Kumar | National Institute of Technology Silchar |
| Mishra, Rajiv Kumar | National Institute of Technology Rourkela, India |
| Chakraborty, Sudipta | NIT Silchar |
Keywords: Robust adaptive control, Networked control systems, Linear systems
Abstract: This paper addresses the leader-follower consensus (LFC) problems of multi-agent systems (MASs) having perturbed double integrator dynamics connected over a digraph. We consider perturbations in the agent dynamics to be bounded by unknown upper bounds. Integral sliding mode (ISM) control is designed to achieve robust tracking performance. We design a backstepping control to reach consensus in almost fixed time, free from singularity. A barrier function-based approach is employed to design the ISM control without information on the bound of the perturbation. We incorporate a positive definite barrier function and discuss the existence of a practical sliding mode (PSM) for the proposed scheme. We provide a simulation example to validate the proposed method.
|
| |
| 12:10-12:30, Paper WeBT2.6 | |
>H_{infinity} Control Design Using Gradient-Descent Method |
|
| Kumar, Amit | Indraprastha Institute of Information Technology |
| Chanekar, Prasad Vilas | Indraprastha Institute of Information Technology |
Keywords: Robust control, Linear systems, Optimization
Abstract: The performance of an engineering system is affected due to the effect of model uncertainties and unknown external disturbances. H_{infinity}-controllers attenuate the worst-case effect of disturbances on the system's performance. In this paper we propose a novel formulation for the full-state feedback H_{infinity}-controller design problem using semi-definite programming duality. We then propose an iterative gradient-based procedure which solves convex optimization problems at each iteration to compute an optimized controller. We also prove that if the procedure converges, then it converges to a stationary solution. Finally we show the utility of the proposed formulation and solution procedure through various practical examples.
|
| |
| WeBT3 |
Seminar Hall 2 |
| Machine Learning and Control I |
Regular Session |
| |
| 10:30-10:50, Paper WeBT3.1 | |
>A Comparative Study of ML and DL Classifiers for Land Cover Classifications |
|
| Jaglan, Satyender | National Institute of Technology Kurukshetra |
| Pal, Mahesh | National Institute of Technology Kurukshetra |
| Dewan, Lillie | NIT Kurukshetra |
Keywords: Machine learning, Neural networks, Pattern recognition and classification
Abstract: Multispectral imaging (MSI) has emerged as a powerful tool for detailed land use and land cover classification, leveraging spectral information across multiple wavelength bands. This study presents a comparative analysis of four advanced classifiers: XGBoost, Multi-Layer Perceptron (MLP), DenseNet, and MLP with a global filter (GMLP), applied to multispectral data from the Central State Farm in Hissar, India. The dataset, covering eight land cover classes includes four spectral bands: Red, Green, Blue, and Infrared. Our findings reveal that the MLP with a global filter achieved the highest classification accuracy of 99.77%, demonstrating superior performance in capturing the intricate spectral-spatial relationships in MSI data. DenseNet also performed exceptionally well, with an accuracy of 99.45%. XGBoost, while being the most computationally efficient, achieved an accuracy of 94.59%, highlighting its robustness. This comparative study underscores the importance of selecting appropriate classifiers for MSI applications, with the GMLP showing the greatest promise for accurate land cover classification.
|
| |
| 10:50-11:10, Paper WeBT3.2 | |
>Machine Learning Based Explainable Fault Detection of Vacuum Pumps in PV Solar Cell Manufacturing |
|
| Narayanan, Ajay | Reliance Industries Limited |
| Kulkarni, Kedar | Reliance Industries Ltd |
| Riswan, Fery | REC Solar Pte Ltd |
| Salehan, Azzrul | REC Solar Pte Ltd |
Keywords: Machine learning, Fault detection/accomodation, Manufacturing systems
Abstract: Pump failures in industrial settings can lead to significant operational disruptions and financial losses.This paper presents a data-driven approach for explainable fault detection in a vacuum pump operating within a chemical vapor deposition (CVD) unit at a Photovoltaic (PV) solar cell manufacturing facility. A comprehensive dataset was collected and several machine learning based classification models were evaluated for their ability to predict pump failures based on the pump sensor data. All models demonstrated exceptional performance, achieving F1-scores ranging from 0.96 to 0.98. Furthermore, explainability analysis on the best model revealed that temperature-related features were the most influential in predicting faults. Additionally, counterfactual analysis provided actionable recommendations to adjust operational parameters in order to mitigate future failures. These findings highlight the potential of machine learning for proactive maintenance and optimization in industrial settings.
|
| |
| 11:10-11:30, Paper WeBT3.3 | |
>BiRoDiff: Diffusion Policies for Bipedal Robot Locomotion on Unseen Terrains |
|
| Gvs, Mothish | Indian Institute of Science |
| Tayal, Manan | Indian Institute of Science, Bengaluru |
| Kolathaya, Shishir | Indian Institute of Science |
Keywords: Intelligent systems, Neural networks, Machine learning
Abstract: Locomotion on unknown terrains is essential for bipedal robots to handle novel real-world challenges, thus expanding their utility in disaster response and exploration. In this work, we introduce a lightweight framework that learns a single walking controller that yields locomotion on multiple terrains. We have designed a real-time robot controller based on diffusion models, which not only captures multiple behaviours with different velocities in a single policy but also generalizes well for unseen terrains. Our controller learns with offline data, which is better than online learning in aspects like scalability, simplicity in training scheme etc. We have designed and implemented a diffusion model-based policy controller in simulation on our custom-made Bipedal Robot model named Stoch BiRo. We have demonstrated its generalization capability and high frequency control step generation relative to typical generative models, which require huge onboarding compute.
|
| |
| 11:30-11:50, Paper WeBT3.4 | |
>Complete Task Learning from a Single Human Demonstration by Identifying Fundamental Sub-Tasks for Robot Manipulator Execution |
|
| Maurya, Anurag | Indian Institute of Technology, Mandi |
| Sharma, Dharmendra | Indian Institute of Technology Mandi |
| Sharma, Archit | Indian Institute of Technology Mandi |
| Thakur, Peeyush | Indian Institute of Technology Mandi H.P |
| Gupta, Sandeep | Indian Institute of Technology Kanpur |
| Dhar, Narendra Kumar | IIT Mandi |
| Behera, Laxmidhar | Indian Institute of Technology Kanpur |
Keywords: Neural networks, Learning, Vision-based control
Abstract: This work tackles the problem of teaching robot manipulators to carry out intricate tasks independently by observing human demonstrations. By breaking down the whole task into smaller basic sub-tasks, a reliable model is developed by associating human actions with equivalent robotic ones more easily so that manipulator actions can be programmed efficiently. To enable real-time sub-task level identification, vision transformer (ViViT), TimeSformer, and VideoMAE models are used separately as encoder architecture and trained using video data to anticipate the sub-task levels. These models are compared, and the accuracy aggregated on different tasks is 64.36% (ViViT), 71.26% (TimeSformer) and 81.03% (VideoMAE). The identified sub-tasks are executed by a robot manipulator using the trajectories learned through the dynamic movement primitives (DMP). Real-time experiments show that this approach greatly enhances the robot’s ability to reliably and precisely reproduce complex tasks. The proposed solution also emphasizes how flexible the system is to various task modifications and how it can be used in multiple healthcare, home, and industrial settings. The novelty lies in the notion of finding semantic connections between primitive sequences from video data.
|
| |
| 11:50-12:10, Paper WeBT3.5 | |
>Active Learning Using One Class SVM for Fault Detection and Diagnosis of Process Systems |
|
| Hambarde, Piyush | Indian Institute of Technology Jodhpur |
| Mandpe, Ankit | Indian Institute of Technology Jodhpur |
| Varanasi, Santhosh Kumar | Indian Institute of Technology Jodhpur |
Keywords: Fault detection/accomodation, Machine learning, Pattern recognition and classification
Abstract: Today's process industries are investing heavily in the field of process safety and efficiency so as to ensure that the plant is running at the optimal conditions and the overall profit is maximized. Early and fast detection and diagnosis of faults is one of the way of ensuring this objective. Owing to better and reliable data acquiring systems and/ or due to boom of Artificial Intelligence in the era of Industry 4.0, researchers are looking at various ways of incorporating different Machine Learning techniques for development of efficient fault detection and diagnosis methods. As it is not feasible to train a model to detect all possible faults that may arise in a plant, it is required to have a methodology, that is capable enough to learn about any new faults that might occur in the due course of operating a process and is the main objective of this paper. To achieve this objective, a two stage approach wherein, in the first stage, an unsupervised learning strategy using one-class SVM is considered in this paper to detect whether a fault is existing or a new one. In the second stage, a multi-class classifier using Neural networks like Extreme learning machine (ELM) has been employed to detect the nature of fault. The efficacy of the proposed method is demonstrated by considering a Tennessee Eastman process and the results are compared with existing methods.
|
| |
| 12:10-12:30, Paper WeBT3.6 | |
>Transfer Learning for Iterative Learning Control Using Gaussian Processes |
|
| Tangirala, Arun K. | Indian Institute of Technology Madras |
| Muthiah, Letchumanan | Indian Institute of Technology Madras |
Keywords: Iterative learning control, Learning, Control applications
Abstract: Iterative Learning Control (ILC) is a technique used to improve the transient tracking performance of a repetitively operated system, such as a batch process. However, the assumption of iteration invariance of the process conditions limits the extension of ILC to similar processes and often demands process modelling and controller design from scratch. Data-driven modelling strategies are highly effective but they suffer from large model development cost during training. This is undesirable since operating conditions of industrial batch processes can switch rapidly to meet the requirements of the market leaving little time and resources for model development. In this paper, we present a novel method of transferring optimally tuned ILC across processes using the modern ideas of transfer learning (TL). Specifically, we develop a Gaussian process model using ideas of forward adaptive transfer learning (FAT) to facilitate knowledge transfer between processes. A major advantage of the proposed method is that optimal ILC designed for current process can be transferred to a new process using only limited training data. The proposed method is demonstrated on the Cott-Batch Reactor, showing substantial promise for the proposed TL-based FAT-GP.
|
| |
| WeBT4 |
Board Room 1 |
| Optimization I |
Regular Session |
| |
| 10:30-10:50, Paper WeBT4.1 | |
>A Distributed Algorithm for Assigning UAVs to Multiple Heterogeneous Positions of Interest |
|
| Gangadari, Varaprasad | Indian Institute of Technology Guwahati |
| Bhawal, Chayan | Indian Institute of Technology Guwahati |
| Kar, Indrani | IIT Guwahati |
Keywords: Optimization algorithms, Multivehicle systems, Cooperative control
Abstract: Unmanned Aerial Vehicles (UAVs) powered by rechargeable batteries face constraints in flight duration and computational capacity. To overcome these limitations, employing multiple UAVs and efficient algorithms becomes essential for handling extensive tasks beyond the capabilities of a single UAV. Many applications, including search and rescue missions, surveillance of tall buildings, and maintenance of heritage sites, often involve diverse and dispersed positions of interest (POIs). When UAVs are stationed across multiple depots and require deployment from these locations, a UAV assignment problem is introduced, considering the heterogeneity of POIs. This assignment problem is modelled as an integer linear programming (ILP) problem aimed at minimizing the total distance travelled by all UAVs. A distributed dual ascent algorithm is proposed to solve the ILP problem efficiently.
|
| |
| 10:50-11:10, Paper WeBT4.2 | |
>On Linear Convergence of PI Consensus Algorithm under the Restricted Secant Inequality |
|
| Chakrabarti, Kushal | Tata Consultancy Services Research |
| Baranwal, Mayank | Indian Institute of Technology, Bombay |
Keywords: Optimization algorithms, Agents-based systems, Stability of nonlinear systems
Abstract: This paper considers solving distributed optimization problems in peer-to-peer multi-agent networks. The network is synchronous and connected. By using the proportional-integral (PI) control strategy, various algorithms with fixed stepsize have been developed. Two notable among them are the PI algorithm and the PI consensus algorithm. Although the PI algorithm has provable linear or exponential convergence without the standard requirement of (strong) convexity, a similar guarantee for the PI consensus algorithm is unavailable. In this paper, using Lyapunov theory, we guarantee exponential convergence of the PI consensus algorithm for global cost functions that satisfy the restricted secant inequality, with rate-matching discretization, without requiring convexity. To accelerate the PI consensus algorithm, we incorporate local pre-conditioning in the form of constant positive definite matrices and numerically validate its efficiency compared to the prominent distributed convex optimization algorithms. Unlike classical pre-conditioning, where only the gradients are multiplied by a pre-conditioner, the proposed pre-conditioning modifies both the gradients and the consensus terms, thereby controlling the effect of the communication graph on the algorithm.
|
| |
| 11:10-11:30, Paper WeBT4.3 | |
>Temperature and Energy Optimization in GPUs: An Illustration |
|
| Das, Rajarshi | Qualcomm India Pvt Ltd |
| Rao, Sachit | International Institute of Information Technology, Bangalore |
Keywords: Optimization, Information technology (IT) systems, Optimization algorithms
Abstract: Optimizing energy consumption of Graphics Processing Units (GPUs) has received considerable interest. Unlike Central Processing Units (CPUs), where energy consumption can be reduced by scaling the core clock frequency dynamically, for a class of workloads, denoted as compute-bound workloads, experimental data indicate that GPUs (the variants considered in this paper) consume the lowest energy at the highest core frequency, thus also leading to the lowest workload runtimes. Operating at the highest frequency, however, leads to a significant increase in the GPU temperature - thus, minimizing both temperature and energy become conflicting objectives. In this work, the simultaneous optimization of energy consumption and GPU temperature is posed as a Multi-Objective Optimization Problem (MOOP) and is solved using the GICOV kernel, borrowed from the medical imaging domain, as an illustration. Experimental data reveal that the speed of the on-board GPU fan and the clock frequency can be used as control variables to solve the MOOP. Pareto-optimal solutions are presented using classical and evolutionary approaches; both yield the same solutions. These solutions show that lower GPU temperatures can be obtained in comparison with default GPU settings. Two additional results are presented: the implementation of a dynamic frequency governor i. as a feedback controller to control temperature alone and ii. as the solution to the Brachistochrone problem to minimize runtime.
|
| |
| 11:30-11:50, Paper WeBT4.4 | |
>Convergence Rates for Stochastic Approximation: Biased Noise with Unbounded Variance, and Applications |
|
| Karandikar, Rajeeva | Chennai Mathematical Institute |
| Vidyasagar, Mathukumalli | Indian Institute of Technology |
Keywords: Statistical learning, Optimization algorithms, Machine learning
Abstract: In this paper, we study the convergence properties of the Stochastic Gradient Descent (SGD) method for finding a stationary point of a given C^1 objective function J(.). which is not required to be convex. If J(.) satisfies Polyak-Lojasiewicz (PL) condition, we derive estimates on the rate of convergence of the function value sequence to its limit. Using these results, we show that for functions satisfying the PL property, the convergence rate of SGD is the same as the best-possible rate for convex functions. While some results along these lines have been published in the past, our contributions contain two distinct improvements. First, the assumptions on the stochastic gradient are more general than elsewhere, and second, our convergence is almost sure, and not in expectation.
|
| |
| 11:50-12:10, Paper WeBT4.5 | |
>Accelerating the Multi-Objective Optimization of Crystallization Process Using High Fidelity Population Balance Model under NUMBA CUDA Environment |
|
| Rangavajhala, Anirudh | Indian Institute of Technology Hyderabad |
| Tadepalli, Aswitha | Indian Institute of Technology Hyderabad |
| Mitra, Kishalay | Indian Institute of Technology Hyderabad |
Keywords: Optimization, Optimization algorithms, Modeling and simulation
Abstract: Crystallization is one of the majorly used separation techniques owing to its high efficiency and ease of operation. The optimization of the crystallization process involves the high-fidelity population balance models, thereby significantly increasing the computational complexity. This study, therefore, explores the acceleration of the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) using NUMBA CUDA for multi-objective optimization for crystallization of ortho-amino benzoic acid. By harnessing GPU parallel processing, significant reductions in computation time are achieved, particularly for large populations and complex optimization problems. The GPU-accelerated NSGA-II algorithm in the optimization of ortho amino benzoic acid crystallization demonstrates significant performance boost by reducing the time by 122 times without any reduction in the solution quality. This research, therefore, underscores the potential of GPU acceleration in evolutionary algorithms and sets the stage for further advancements in high-performance parallelized optimization techniques, promising significant improvements in real-time optimization scenarios.
|
| |
| 12:10-12:30, Paper WeBT4.6 | |
>Mathematical Programming Formulations for Computing Nash Equilibrium of Stochastic Games |
|
| Verma, Ayush | Indian Institute of Technology Delhi |
| Singh, Vikas Vikram | Indian Institute of Technology Delhi |
| Palkar, Prashant | Indian Institute of Technology Delhi |
Keywords: Markov processes, Optimization algorithms, Stochastic systems
Abstract: In this paper, we propose a trilinear optimization formulation and four mixed integer bilinear formulations to compute the Nash equilibria of a 2-player discounted stochastic game. The trilinear optimization formulation is derived from an existing nonlinear programming problem whose global minimum gives a Nash equilibrium of the discounted stochastic game. The mixed integer bilinear formulations use the optimality condition of a best response Markov decision problem of a player for a fixed strategy of the other player. We compare the performance of our formulations with the existing nonlinear programming formulation. Numerical experiments show that our trilinear formulation outperforms the nonlinear programming formulation as well as the mixed integer bilinear formulations.
|
| |
| WeDT1 |
Auditorium |
| Identification and Estimation I |
Regular Session |
| |
| 15:15-15:35, Paper WeDT1.1 | |
>Minimum Operator Based Discrete-Time Nonlinear Observer Design for the State of Charge Estimation of Li-Ion Battery |
|
| Shiwangi, Shilpi | Indian Institute of Technology (BHU) Varanasi |
| Pandey, Sunidhi | IIT(BHU) Varanasi, India |
| Ghosh, Sandip | IIT BHU, Varanasi |
| Kamal, Shyam | Indian Institute of Technology (BHU), Varanasi |
| Pandey, J C | IIT (BHU) Varanasi |
| Taslima, Eram | IIt BHU (Varanasi) |
Keywords: Estimation, Observers for nonlinear systems, Control applications
Abstract: This paper introduces a novel minimum operator-based discrete-time nonlinear observer for estimating the online state of charge (SoC) of Li-ion batteries. This observer ensures finite-time estimation error convergence. The battery’s model employs a second-order resistor-capacitor (2-RC) equivalent circuit, where the parameters are determined via a least-squares fitting method and treated as nonlinear functions. The Lyapunov direct method is used to demonstrate the finite-time convergence of the estimation error in discrete time. The proposed method’s performance is validated under hybrid pulse power characterization (HPPC) and urban dynamometer driving schedule (UDDS) conditions. Simulation and experimental results show that it provides better estimation accuracy and tracking capability than existing discrete-time nonlinear observers.
|
| |
| 15:35-15:55, Paper WeDT1.2 | |
>DATA-DRIVEN MODEL for ESTIMATION of SoH and INTERNAL RESISTANCE of LITHIUM ION BATTERY BASED on DEEP NEURAL NETWORK |
|
| Viswanathan, Mehanthika | Anna University |
| D, Rishi | Madras Institute of Technology, Anna University |
| S, Sutha | Madras Institute of Technology Campus, Anna University Chennai |
| Natarajan, Pappa | Professor, Department of Instrumentation Engineering, MIT Campu |
Keywords: Estimation, Neural networks, Automotive
Abstract: Accurate estimation of State of Health (SoH) and Internal Resistance (IR) is essential for optimized performance and longevity, particularly in the new-era of electric vehicles and renewable energy storage devices. The traditional methods struggle with battery data complexity in dynamic situations to enhance the above parameters. This paper presents a data-driven model with hybrid architecture that combines CNN and LSTM models which revolutionize the battery management and focusing on the SoH and IR estimation in dynamic environments, CNN-LSTM models excel by extracting spatial features and learning temporal dependencies within the battery datasets. This increases accuracy and enables real-time monitoring, diagnosis, and optimization of battery health. This paper also shows the execution of LSTM with and without sampling of the datasets for efficient runtime and produces accurate results, this marks a significant leap in the battery management and enabling smarter resource utilization for sustainable energy solutions.
|
| |
| 15:55-16:15, Paper WeDT1.3 | |
>Parameter Identification and Control of Integrating Time Delay Systems |
|
| Pandey, Saurabh | Rajiv Gandhi Institute of Petroleum Technology, Jais |
| Majhi, Somanath | Indian Institute of Technology Guwahati |
Keywords: Identification, Linear systems, Modeling and simulation
Abstract: This paper presents the ideal relay-based modeling and control schemes for a class of time-invariant linear systems exhibiting integrating and time delay dynamics. Based on the important parameters in induced system response, a set of explicit equations is derived for evaluation of parameters associated with an assumed integrating plus time delay transfer function model. Under the adverse effects of sensor noise, a best-fit output is recovered by using a curve fitting method based on a Fourier series approach. The range of normalized dead time to which conditions for existence of limit cycles in an integrating and time delay systems is further enhanced by incorporating a feedback controller. For efficient and robust control design, a combination of feedforward proportional-integral (PI) and feedback proportional (P) controllers is utilized. Tuning guidelines for each of the controller gains are established through an integral of time-weighted error criterion. Illustrative examples are studied to validate the proposed algorithms for parameter identification and controller design. Furthermore, the efficacy of methods proposed is shown by using frequency and step response plots.
|
| |
| 16:15-16:35, Paper WeDT1.4 | |
>Centralized Fusion of Radar and IRST on an Aircraft |
|
| Ailneni, Sanketh | National Aerospace Laboratories |
| Kashyap, Sudesh Kumar | National Aerospace Laboratories |
| Sinha, Nandan kumar | IIT MADRAS |
Keywords: Sensor fusion, Estimation, Kalman filtering
Abstract: This paper presents a centralized sensor fusion approach for radar and an infrared search and track (IRST) system mounted on an aircraft. The target motion is modelled in relative kinematics with radar measuring range, azimuth and elevation while IRST measures only target line of sight direction of arrival. A detailed sensor fusion architecture is presented for target tracking and the performance of the same is analyzed in the simulation environment. The sensor operational modes viz. track while scan (TWS) and single target tracking (STT) modes of radar and IRST are analyzed and target tracking performance is presented. Further, Monte Carlo simulations are carried out and root mean squared error analysis has been used to analyze the proposed centralized sensor fusion algorithm.
|
| |
| 16:35-16:55, Paper WeDT1.5 | |
>System-Theoretic Insights into Battery Power Fade Prediction: A Probabilistic Framework |
|
| Brahma, Sarnaduti | Siemens Limited, India |
| Relan, Relan | Siemens Limited, India |
Keywords: Identification, Machine learning, Estimation
Abstract: Accurate estimation of the power fade of a battery is crucial for its optimised usage and extended lifetime. Non-invasive techniques such as Electrochemical Impedance Spectroscopy (EIS) can be combined with machine learning to estimate power fade from externally measurable signals without damaging the battery. Traditional methods in the literature involve complete EIS curves obtained through separate sinusoidal excitations at various frequencies as features for training models, which is computationally intensive and time-consuming. This paper proposes a Gaussian Process Regression-based conformal prediction model for estimating a cell's power fade using EIS. The approach leverages system-theoretic concepts like multisine excitation, transfer function representation and subsequent pole-zero extraction. Multisine excitation allows for simultaneous excitation of relevant frequencies. At the same time, extracting only poles/zeros from EIS curves instead of using the entire EIS reduces the size of the training space, lowering computational resource requirements. Simulation results on a validated cell model demonstrate the accuracy of the proposed method across different temperatures in the presence of measurement noise.
|
| |
| 16:55-17:15, Paper WeDT1.6 | |
>Self-Calibrating Offline Trajectory Estimation Technique for Sporadically Observable Systems |
|
| Buriuly, Somnath | Indian Institute of Technology Bombay |
| Vachhani, Leena | Indian Institute of Technology, Bombay |
Keywords: Optimization algorithms, Filtering, Estimation
Abstract: Industrial problems, like pipeline inspection and railway track geometry inspection inside tunnels, require accurate offline estimation of the state trajectories, to determine the infrastructure parameters, say pipeline GPS coordinates, rail-track's attitude and placement, etc. These problems are cursed with loss of observability due to the unavailability of localization data, like GPS, inside pipelines and tunnels. In addition, the states at the start and end of the recorded data are observable but not accurately or entirely known. Furthermore, a deviation is observed in the initial calibration of the Inertial Measurement Unit (IMU) sensors, and hence the recorded data are approximated sensor readings. Since the accuracy of these estimated outcomes impacts the maintenance cost of the infrastructure, it is essential to obtain the best possible estimate from the offline available sensor data. In this work, we propose an optimization-based solution to estimate the entire trajectory and self-calibrate the sensor readings, simultaneously. In particular, we formulate an offline state trajectory estimation problem with self-calibration features for the sensor readings, compute the gradients for the decisions, and propose a Lagrangian-based gradient descent algorithm. The results show improvement in the resulting accuracy of the estimates in the presence of inaccurate readings.
|
| |
| WeDT2 |
Seminar Hall 1 |
| Autonomous Systems I |
Regular Session |
| |
| 15:15-15:35, Paper WeDT2.1 | |
>Nonlinear Control Law for Path Following by UAVs |
|
| Pal, Sayantan | Indian Institute of Technology Kharagpur |
| Hota, Sikha | Indian Institute of Technology Kharagpur |
Keywords: Autonomous systems, Nonholonomic systems, Nonlinear systems
Abstract: This work finds a solution to reduce the convergence time by proposing a nonlinear guidance law for path following problems of Unmanned Air vehicles (UAVs). The guidance law has been derived based on the assumption that a virtual target tracks the desired path accurately and the UAV tries to catch the virtual target, that is, the relative distance between them and the look angle are made to zero. The proposed guidance law provides a generic solution which is applicable for any path-following problem. Comparative results showcase the efficacy of the proposed guidance law in terms of convergence time.
|
| |
| 15:35-15:55, Paper WeDT2.2 | |
>Traversal of Unmanned Autonomous Vehicles through Arbitrarily Shaped Openings |
|
| V, Karthickeyan | IIT Bombay |
| Kumar, Shashi Ranjan | Indian Institute of Technology Bombay |
| Arya, Hemendra | IIT Bombay |
Keywords: Autonomous systems, Vision-based control, Computational methods
Abstract: In recent years, deploying unmanned aerial and ground vehicles has become increasingly prevalent across various domains, such as agriculture, defense, and filmmaking, owing to their versatility and efficiency. Despite their widespread use, these robots are still facing challenges in certain scenarios, particularly in navigating through arbitrarily shaped openings. This paper addresses the above challenge by proposing algorithms designed for opening detection and feasibility checks, ensuring safe passage through these complex environments. The core of this work revolves around the contour detection algorithm, developed using OpenCV, and the optimal rectangle extraction algorithm, which significantly enhances computational efficiency and is employed to fit an optimal rectangle within the detected contour. This step is crucial for ensuring the robot's safety by generating a feasible goal point for the point-tracking algorithm to follow. To validate the effectiveness of the proposed algorithms, a comprehensive experimental setup, including a ground robot equipped with various state estimation techniques, was established. Extensive testing and validation were conducted within this experimental framework, demonstrating the robustness and reliability of the algorithms in real-world scenarios. The successful implementation of these algorithms in hardware attests to this research's practical applicability and potential impact.
|
| |
| 15:55-16:15, Paper WeDT2.3 | |
>Emergency Vehicle Collision Avoidance Using LSTM-Based Trajectory Predictions and Collision Cones |
|
| Hindoliya, Rajkumar | Indian Institute of Science Banglore |
| Ghose, Debasish | Indian Institute of Science |
Keywords: Autonomous systems, Vision-based control, Machine learning
Abstract: This paper proposes a novel collision avoidance technique for an Emergency Autonomous Vehicle (EAV) operating in an environment with multiple human entities moving in arbitrary trajectories. The EAV is equipped with a camera and an auditory signalling device. An Inverse Perspective Mapping (IPM) technique is used to estimate the location of the pedestrians via a bounding box location obtained from the camera image. A model to generate a Birds Eye View (BEV) of the EAV-pedestrian environment using Conditional Generative Adversarial Networks (cGAN) is proposed to aid decision-making. The future trajectory of the pedestrians is predicted using an Long short-term memory (LSTM) model. Impending collisions are predicted using the concept of collision cones, and the effect of using the auditory signaling device at different time instances is investigated. Pedestrian datasets are collected in real-life scenarios, and the effectiveness of the proposed techniques is demonstrated via field experiments.
|
| |
| 16:15-16:35, Paper WeDT2.4 | |
>Enhanced A* Algorithm for Mobile Robot Path Planning with Non-Holonomic Constraints |
|
| Kumar, Suraj | Indian Space Research Organization |
| R, Sudheendra | Indian Institute of Science |
| Rallapalli, Aditya | UR Rao Satellite Center |
| Gvp, Bharat Kumar | UR Rao Satellite Centre |
| Lagisetty, Ravi Kumar | Indian Space Research Organization |
Keywords: Autonomous systems, Nonholonomic systems
Abstract: In this paper, a novel method for path planning of mobile robots is proposed, taking into account the non-holonomic turn radius constraints and finite dimensions of the robot. The approach involves rasterizing the environment to generate a 2D map and utilizes an enhanced version of the A* algorithm that incorporates non-holonomic constraints while ensuring collision avoidance. Two new instantiations of the A* algorithm are introduced and tested across various scenarios and environments, with results demonstrating the effectiveness of the proposed method.
|
| |
| 16:35-16:55, Paper WeDT2.5 | |
>Unified Dual-Wheel PID Control for Differential Drive Robot with Distinct Wheel Error Functions |
|
| Naragani, Yogi | UPES Dehradun |
| Gariya, Deepak | UPES Dehradun |
| Singhal, Rahul | University of Petroleum and Energy Studies (UPES) |
Keywords: PID control, Control applications, Autonomous systems
Abstract: The PID controller is the most versatile controller due to its ease of implementation. However, multi-input control design with PID controller is a challenging task. This paper presents the unified PID implementation for the right and left wheels of the differential drive robot. The distinct error function was considered for each wheel utilizing the robot’s kinematic relationship. The simulation result of the proposed unified dual wheel PID (UDW-PID) controller presented for the straight, circular and sine path. The control action generated for the respective wheel resulted in smoother and quicker convergence towards the path, and afterwards stable behaviour over the path.
|
| |
| 16:55-17:15, Paper WeDT2.6 | |
>Formation Flying for Fixed Wing UAVs with Obstacle Avoidance through Narrow Gaps |
|
| Antony, Anish | VSSC, ISRO |
| Kumar, Shashi Ranjan | Indian Institute of Technology Bombay |
| Mukherjee, Dwaipayan | Indian Institute of Technology Bombay |
Keywords: Aerospace, Control applications, Autonomous systems
Abstract: A formation control algorithm based on artificial potential fields for fixed-wing unmanned aerial vehicles to traverse through environments filled with non-cooperative static obstacles is proposed in this paper. The spherical formation is created with a virtual leader at the center of the sphere and other UAVs positioned on the circumference at a predefined radius. The formation is created using a combination of leader-follower and behavioural formation structure concepts combined with an improved artificial potential field approach. The agents are equipped with obstacle avoidance capabilities through the rotational artificial potential field method. The common issues of local minima entrapment can be avoided with this concept. The paper introduces an algorithm that will empower the formation to travel through narrow passages using dynamic formation reconfiguration. When the obstacles are near, an algorithm to generate a convex hull of obstacles is also proposed in this work. The efficacy of the proposed algorithms is demonstrated through MATLAB simulations.
|
| |
| WeDT3 |
Seminar Hall 2 |
| Machine Learning and Control II |
Regular Session |
| |
| 15:15-15:35, Paper WeDT3.1 | |
>Reinforcement Learning with Quasi-Hyperbolic Discounting (Extended Abstract) |
|
| S R, Eshwar | Indian Institute of Science |
| Motwani, Mayank | Indian Institute of Technology, Bombay |
| Roy, Nibedita | Indian Institute of Science |
| Thoppe, Gugan | Indian Institute of Science |
Keywords: Iterative learning control, Agents-based systems, Optimization algorithms
Abstract: Human decision-making is better modeled via discounting schemes such as quasi-hyperbolic and hyperbolic than classical ones such as exponential or average reward. In a recent work, we initiated the study of reinforcement learning with Quasi-Hyperbolic (QH) discounting and proposed the first model-free algorithm for finding a type of policy called the Markov perfect equilibrium. This paper provides a summary of that algorithm and our other findings on its convergence.
|
| |
| 15:35-15:55, Paper WeDT3.2 | |
>Physics Informed Deep Operator Networks for Baculovirus-Insect Cell System |
|
| Masampally, Vishnu Swaroopji | Tata Consultancy Services Limited |
| Sharma, Surbhi | Indian Institute of Technology, Hyderabad |
| Giri, Lopamudra | Indian Institute of Technology, Hyderabad |
| Mitra, Kishalay | Indian Institute of Technology Hyderabad |
Keywords: Neural networks, Machine learning, Biological systems
Abstract: The Baculovirus expression vector system (BEVS) is extensively used for producing recombinant proteins, virus-like particles (VLPs), and vaccines. Optimizing and controlling this process requires a robust model tuned with experimental data. A physics-based mathematical model for BEVS has been developed and validated through experiments. This model has been used to create a Physics-Informed Deep Operator Network (PI-DeepONet) model. Given that the mathematical model involves stiff ordinary differential equations (ODEs), the Stiff-PINN formulation, a variant of Physics-Informed Neural Networks (PINNs), is utilized to construct the loss function. The PI-DeepONet model effectively predicts process dynamics across various initial conditions, whereas the Stiff-PINN model is limited to the initial condition for which it was trained. Once developed, PI-DeepONet models can replace standard PINN models, enabling the optimization and control of BEVS across diverse initial conditions.
|
| |
| 15:55-16:15, Paper WeDT3.3 | |
>Reinforcement Learning Based Attitude Control of Quadcopter |
|
| K, Sneha Mohan | College of Engineering Trivandrum |
| Rosaline, Arlene Davidson | College of Engineering Trivandrum |
| Mishra, Deepak | IIST Trivandrum |
Keywords: Control applications, Machine learning, Flight control
Abstract: This research work explores the application of reinforcement learning (RL) for the attitude control of a quadcopter in a pick and place scenario. The primary objective is to achieve stable and precise control over the quadcopter’s pitch, roll and yaw angles. Traditional control systems for quadcopters often rely on predefined models and PID controllers, which may struggle with dynamic environments and uncertainties. The Actor-Critic method, a powerful RL technique, is employed to train that can effectively stabilize the quadcopter’s orientation. The Actor Critic method comprises two main components: the Actor, which proposes actions based on current policy and the Critic, which evaluates these actions by computing the value function. This dual mechanism enables the quadcopter to adaptively improve its control policy through continuous interaction with the environment. The training process involves the quadcopter performing a series of attitude variations, receiving rewards based on its performance, and updating its policy accordingly. Through continuous interaction with the environment, the actor learns to improve its policy, while the critic provides valuable feedback to guide this learning process. An Reinforcement Learning (RL) based approach for quadcopter attitude control with a focus on pick and place application.
|
| |
| 16:15-16:35, Paper WeDT3.4 | |
>SANGO: Socially Aware Navigation through Grouped Obstacles |
|
| Malladi, Rahath | Plaksha University |
| Harsh, Amol | Plaksha University |
| Sangwan, Arshia | Plaksha University |
| Chauhan, Sunita | Plaksha University, India |
| Manjanna, Sandeep | Plaksha University |
Keywords: Learning, Autonomous systems, Machine learning
Abstract: This paper introduces SANGO (Socially Aware Navigation through Grouped Obstacles), a novel method that ensures socially appropriate behaviour by dynamically grouping obstacles and adhering to social norms. Using deep reinforcement learning, SANGO trains agents to navigate complex environments leveraging the DBSCAN algorithm for obstacle clustering and Proximal Policy Optimization (PPO) for path planning. The proposed approach improves safety and social compliance by maintaining appropriate distances and reducing collision rates. Extensive experiments conducted in custom simulation environments demonstrate SANGO's superior performance in significantly reducing discomfort (by up to 83.5%), reducing collision rates (by up to 29.4%), and achieving higher successful navigation in dynamic and crowded scenarios. These findings highlight the potential of SANGO for real-world applications, paving the way for advanced socially adept robotic navigation systems.
|
| |
| 16:35-16:55, Paper WeDT3.5 | |
>Reinforcement Learning Based Persistent Surveillance on Graphs under Energy Constraints |
|
| Hoysal, Nishchal | Indian Institute of Science |
| S, Aravind | Indian Institute of Science, Bangalore |
| Jayanti, Lalit | Indian Institute of Technology Madras |
| Tallapragada, Pavankumar | Indian Institute of Science |
Keywords: Learning, Intelligent systems, Control applications
Abstract: In this work, we consider the problem of optimally routing a single agent on a graph for persistent surveillance and reporting under energy constraints. We aim to minimize a combination of weighted graph idleness, based on node priorities, and the time elapsed since the agent's last report. This requires repeatedly solving `similar' computationally challenging optimization problems. To this end, we adopt a rolling horizon approach and use a deep Q-network (D3QN) based algorithm to learn policies that, given the graph state and the agent's location, decide the agent's next node to visit, while guaranteeing the satisfaction of energy constraints. Through exhaustive simulations and comparisons against an integer programming (IP) solver, we demonstrate near-optimality of our approach. Our method also provides several orders of magnitude reduction in computation time over the IP solver. We also compare our approach with five other frameworks, namely, Q learning based Black Box Learning Agent (BBLA), Entropy Maximized Patrolling (EMP), Travelling salesman problem (TSP) heuristic, Concurrent Bayesian Learning Strategy (CBLS), a graph attention (GAT) based learning method.
|
| |
| 16:55-17:15, Paper WeDT3.6 | |
>Accurate Detection of Bright Spots in Solar Photovoltaic Panels |
|
| S Nair, Rahul | Reliance Industries Limited |
| Thirwani, Anand | Reliance Industries Limited |
| Kulkarni, Kedar | Reliance Industries Ltd |
Keywords: Machine learning, Fault detection/accomodation, Optimization
Abstract: For optimal performance and safety of photovoltaic (PV) solar panels, early detection of manufacturing defects is very critical. A major concern with PV solar panels is the presence of defects on its surface that show up as "bright spots" in their electro luminescence (EL) images. This defect can lead to malfunctions due to uncontrolled energy production, potentially resulting in short circuits, elevated temperatures, and even fire hazards. In this paper, we propose a robust machine learning (ML) based approach to accurately detect bright spots by optimally splitting the EL images of PV solar panels and engineering novel discriminative features. The proposed method's effectiveness was validated with an impressive F1 Score of 0.96 on real EL images captured directly from a functioning PV solar panel manufacturing unit. This signifies its real-world applicability for reliable bright spot detection in PV solar panels.
|