arXiv subjects

Dhruv Sirohi

Publications and source records attributed to Dhruv Sirohi.

2 recordsLinked to original sources

arXiv2024

Bayesian Optimization Framework for Efficient Fleet Design in Autonomous Multi-Robot Exploration

This study addresses the challenge of fleet design optimization in the context of heterogeneous multi-robot fleets, aiming to obtain feasible designs that balance performance and costs. In the domain of autonomous multi-robot exploration, reinforcement learning agents play a central role, offering adaptability to complex terrains and facilitating collaboration among robots. However, modifying the fleet composition results in changes in the learned behavior, and training multi-robot systems using multi-agent reinforcement learning is expensive. Therefore, an exhaustive evaluation of each potential fleet design is infeasible. To tackle these hurdles, we introduce Bayesian Optimization for Fleet Design (BOFD), a framework leveraging multi-objective Bayesian Optimization to explore fleets on the Pareto front of performance and cost while accounting for uncertainty in the design space. Moreover, we establish a sub-linear bound for cumulative regret, supporting BOFD's robustness and efficacy. Extensive benchmark experiments in synthetic and simulated environments demonstrate the superiority of our framework over state-of-the-art methods, achieving efficient fleet designs with minimal fleet evaluations.

cs.RO↗

arXiv2023

Novel applications of Generative Adversarial Networks (GANs) in the analysis of ultrafast electron diffraction (UED) images

Inferring transient molecular structural dynamics from diffraction data is an ambiguous task that often requires different approximation methods. In this paper we present an attempt to tackle this problem using machine learning. While most recent applications of machine learning for the analysis of diffraction images apply only a single neural network to an experimental dataset and train it on the task of prediction, our approach utilizes an additional generator network trained on both synthetic data and experimental data. Our network converts experimental data into idealized diffraction patterns from which information is extracted via a convolutional neural network (CNN) trained on synthetic data only. We validate this approach on ultrafast electron diffraction (UED) data of bismuth samples undergoing thermalization upon excitation via 800 nm laser pulses. The network was able to predict transient temperatures with a deviation of less than 6% from analytically estimated values. Notably, this performance was achieved on a dataset of 408 images only. We believe employing this network in experimental settings where high volumes of visual data are collected, such as beam lines, could provide insights into the structural dynamics of different samples.

physics.chem-ph↗