Paper 13118-65
Paper 13118-65
Actor-critic reinforcement learning for autonomous navigation in a microscopic world
19 August 2024 • 5:30 PM - 7:00 PM PDT | Conv. Ctr. Exhibit Hall A
Abstract
The report exhibits autonomous navigation of a microparticle in noisy environment aided by real time detection and machine learning enabled feedback mechanism. Specifically, we use Actor-Critic Reinforcement Learning (ACRL) algorithm for navigation of a microparticle to a specific target location under the inevitable influence of Brownian motion as well as flow fields. Furthermore, we investigate a multi-agent reinforcement learning model (MARL) for navigation of multi particle systems where specific target locations and interparticle separation are provided as sensory inputs. Our study demonstrates that, the model trained on a noisy environment can learn effective policy for navigation, reaching various target states after limited number of exploration episodes.
Presenter
Diptabrata Paul
Univ. Leipzig (Germany)
I obtained my Masters and PhD degree in 2023 from Indian Institute for Science Education and Research (IISER) Pune, India working on spin-orbit and opto-thermal interaction with focused optical beams. During my doctoral degree I was also an active member of a student chapter called IISER Pune Optics league and later being a president of the chapter organized weekly seminar on various topics related to light-matter interaction. During my PhD, in 2022 I was also a visiting research student at Molecular Nanophotonics group (with Prof. Frank Cichos), Peter Debye Institute for Soft Matter Physics at Faculty of Physics and Earth Sciences, Universität Leipzig. Later after completion of my doctoral degree I joined the same group as a postdoctoral research associate, where I am currently working on various aspects of opto-thermal control, manipulation and advanced imaging techniques.