Recognized as the most relevant, innovative, impactful and commercially viable B.Tech. final year project
Sloshing refers to the motion of the free liquid surface inside its container. It is a complex nonlinear dynamical phenomenon that has a substantial impact on the fluid system’s stability. It affects various engineering systems and processes such as liquid storage tanks, liquid rocket fuel tanks, molten metal handling in steel plants, robotic handling of liquids, etc. We aim to solve the problem of minimizing slosh in Automated Ground Vehicle (AGV) payloads, i.e, stabilize the free surface of a liquid inside a container placed as payload on an AGV, while the AGV traverses along specified paths in a 2-D plane. For this purpose, a Deep Reinforcement Learning (DRL) framework will be designed to tune a robust controller to control the prototype AGV and move it to a destination point along desired 2-D paths while minimizing the slosh of the payload liquid as well as minimizing the time taken to reach the destination.
@thesis{bachelorsthesis,title={{Reinforcement Learning Based Stabilization of Liquid Surface in Ground Vehicle payloads}},author={Bithel, Kshitij and Dixit, Keshav},year={2022},}
Deep Reinforcement Learning based Super Twisting Controller for Liquid Slosh Control Problem
Ashish Shakya, Kshitij Bithel, Gopinath. Pillai, and Soham Chakrabarty
Deep Reinforcement Learning (DRL) based parameter optimization of super twisting control (STC) for the liquid slosh control problem in a moving vehicle is proposed in this paper. The slosh control problem, including the vehicle dynamics, represents an under-actuated nonlinear dynamical system. The slosh phenomenon is modeled by a simple pendulum on a cart and STC had been designed for the system when the vehicle motion is in a straight line. In this paper, a DRL framework is designed for the first time to tune the STC parameters in order to deliver near optimal performance. The effectiveness of this proposed learning-based approach for STC design for the slosh control problem is validated in a Python simulation environment and compared to the simple STC design without the learning.
@article{2022drlbasedsloshcontrol,title={{Deep Reinforcement Learning based Super Twisting Controller for Liquid Slosh Control Problem}},author={Shakya, Ashish and Bithel, Kshitij and Pillai, Gopinath. and Chakrabarty, Soham},journal={IFAC-PapersOnLine},volume={55},number={1},pages={734--739},year={2022},doi={10.1016/j.ifacol.2022.04.120},}
Reinforcement Learning Based Stabilization of Liquid Surface in Ground Vehicle payloads2022
