Research Column

We, humans, are continuously trying to push our limitations by exploring uncharted territories in the universe. With the explosion of robotics applications, I believe we may expand our presence (remotely) to conduct research or explore extraterrestrial bodies, farm fields, deep oceans, mining fields and more. However, robots that use fewer computing resources and energy while predicting the optimal path in an uncertain or complex world are ideal for the aforementioned missions or tasks. My current research is centred on exploring the implications of this topic with Dr. Chinmay Desai and Dr. Pooja Shah. We are trying to develop a computationally efficient and lightweight inverse-kinematics algorithm that computes the bounds of a kinematic state based on the uncertainty in wheel-terrain contact. Instead of solving the whole kinematic rover-terrain settling, the essential principle of our new technique is to rapidly calculate the vehicle configuration bounds. Knowing the bounds of some important states, the suggested technique may efficiently generate a conservative assessment of the system-terrain clearance, rover attitude, and suspension angles in a closed form.

Optimization of Rocker-Bogie Mechanism using Heuristic Approaches

The objective of this research was to improve the design of the wheel suspension system of a planetary rover based on well-defined mobility criteria using optimization paradigms in conjunction with a rigorously developed objective function that takes into consideration the stimulatory data for the rover's associated commissions. With the advent of computational technologies, predictive analysis, and the multi-objective use of optimization, seven algorithms were leveraged.
Supervisors: Dr. Pooja Shah and Dr. Paresh Gujarati

A comparative study on the modern deep learning architectures for predicting nutritional deficiency in rice plants

Plant nutrient insufficiency identification is the process of detecting the number of nutrients deficient in the soil. In this project, we have used five famous deep learning architectures namely, ResNets, DenseNets, MobileNets, VGG-16, and SqueezeNet to identify insufficiency of nutrients like Nitrogen, Calcium and Boron by this Non-destructive method. We created a hydroponic experiment in which we gathered images for three different types of nutrient deficiencies (N, K, and Ca) and compared them to rice plants growing in full nutrition, for a total of four classes.
Supervisors: Asst. Prof. Kirti Bardhan and Dr. Pooja Shah.