PhD Thesis: Manipulators and Manipulation in High Dimensional Spaces

Advisor: Prof. Emo Todorov (UW-CSE), Prof. Sergey Levine (UC-Berkeley)

Hand manipulation is one of the most complex form of biological movements. Despite its significance in multiple fields such as biomechanics, neuroscience, robotics and graphics, our understanding of dexterous manipulation is quite superficial and far from being reproducible. The unique capabilities of the human hand have long inspired roboticist in their pursuit to develop manipulators with similar dexterity. Simple and isolated tasks such as grasping can of course be accomplished by simpler devices. Nevertheless, if robots are to perform a wider range of tasks in less structured environments than what is currently possible, they are likely to need manipulators approaching human levels of dexterity. The primary goal of this thesis is to realize dynamic dexterous manipulation on physical hardware. The thesis makes following contributions towards this goal Design and development of “ADROIT Manipulation Platform” – a 28 degrees of freedom arm-hand system capable of hosting dexterous dynamics manipulation, thanks to the custom designed fast, strong and compliant pneumatic actuation system. Behavior synthesis techniques – that are capable of synthesizing the details of dynamic dexterous manipulation in high dimensional manipulators – and scalable approaches that bridge the wide divide between what’s possible in simulation and what works on the physical hardware under real world conditions.

Thesis: (source1), (source2)

Master's Thesis: Fuzzy genetic Algorithms(FGA)

Advisor: Prof. Debjani Chakraborty, Dept. of Mathematics, IIT Kharagpur

FGA proposes a unique solution strategy for optimizing fuzzy multi objective problems. Extrapolating standard ZDT test functions to fuzzy domains, new benchmark test functions (FZDT) for fuzzy optimization problems have been proposed.

Thesis (pdf), Paper(pdf)

Under-grad Thesis: New Genetic Alogrithm based multi-objective optimization algorithm(NMGA)

Advisor: Prof. Nirupam Chakraborty, Head of Dept. of Metallurgical & Materials Engineering, IIT-Kharagpur

The Algorithm works on a neighborhood concept in the functional space, utilizes the ideas on weak dominance and ranking and uses its own procedures for population sizing. NMGA performs well against standard test functions, and when applied to a real-life data of integrated steel plant, it outperformed other multi-objective evolutionary algorithm.

Thesis (pdf)