Machines

	SAP-assembly line integration Developed a contextual extraction system to understand the generated SAP error codes and relay the information to the conveyor system of the assembly line using a custom built communication interface. Observed improved traceability of the assemblies produced on the line.
	Vision based burr detection for robotic fettling process Developed a vision based burr detection system which alerts the operator if the burr size crosses a predefined threshold value. This prevents any further damage to the abrasive cutter setup.
	Vision based petrol tank model identification system Developed an end-to-end vision based flexible model identification system. The state-of-the art resnet models from Pytorch are used for identification of the petrol tanks. The entire solution is developed using Flask, a Python based rapid web application deployment package. The model is made modular to allow for on-line addition of new tank designs. A faster inference time and a high frame rate are achieved by optimising the model through quantisation.
	Vision based weld classification system on the edge Robust vision based classification system for different welding of mufflers. Developed an in-house low cost deep learning based machine vision solution using an edge device and a webcamera. Pytorch's mobilenet_v2 improves the frame rate at the cost of classification accuracy between the components when implemented on a resource constraint edge devices. Inference of the serialised model using torch library on C++ further improves the frame rate.

Projects

	Accessible topology optimisation Implementation of the 88 line Matlab function "top88" in C++ using the Eigen library.
	Dual Quaternion Control This project aims to explore the control of a manipulator with dual quaternion representation. Also studies the effects of choice of error to be controlled and its implication on the physical robot.
	The Rising Camel Problem: Trajectory optimisation of parallel manipulators Considering a quadruped robot in its quadruple stance phase as a parallel manipulator, we attempt to solve the energy-optimal control from sitting to a standing posture, also known as the rising camel problem. The idea of powerlifting is also explored.
	Kalman Filter Implementation Implementation of an Extended Kalman Filter (EKF) on the data obtained from a non-linear chemical plant model.
	Parametric Study of Energy Optimal Control of a 2-R Manipulator Implementation of time and energy optimal control on a 2-R planar manipulator and study the effects of geometric and inertial parameters on the control law.
	Design of a 5-bar parallel manipulator Optimal design of a 5-bar parallel manipulator with respect to the Global Conditioning Index (GCI), Measure of Manipulability (MoM) and maximisation of the Safe Working Zone (SWZ) using NSGA-II

Tutorials

	Cart-Pole Problem Revisited: Trajectory Optimisation and more A compilation of a few methods to control the cart-pole system and answer some questions on the way.
	Boid simulation in C++ and OpenGL Simulating the naturally emergent flocking behaviour as proposed by Craig Reynolds in C++ using OpenGL.
	Simulating the game of life in C++ The game of life is an example of how complex behaviour can emerge from a simple set of rules.
	Dzhanibekov effect: Simulations in Fortran vs C++ The idea of this tutorial is (1) To simulate the Dzhanibekov effect (2) Compare the speed and accuracy of simulation in C++ and Fortran.
	Index Theorems Index theorems gives us a qualitative way to understand the vector field of a dynamical system, i.e., to know if there is a possibility of a closed orbit or to know what type of fixed points are possible within a curve etc.
	Flexible rotating body simulation Simulation of the Lagrangian formulation of a flexible rotating beam modeled using Finite Element Method (FEM).

Archive