Hello, I am Akash
I am an enthusiast continuously working towards integrated intelligent systems. The key focus areas include robot system design and integration with sensors like Camera, Radar, GPS etc .
PORTFOLIO COMPUTER VISION EXPERIMENTS
1. Embedded Deep Learning based Semantic Segmentation using TensorFLow Lite
This work explores the task of traffic scene segmentation using convolution neural network. The trained models are passed through 4 compression techniques (without quantization, Float16, Dynamic range, Int8) using TensorFlow Lite. The compression allows the model to achieve the maximum reduction of up to 91 percent with reduced noise.
Keywords — Computer Vision, Semantic Segmentation, Embedded Deep-Learning, Model Compression, Quantization
Tools — Python, TensorFlow, Keras, TensorFlow Lite, Numpy, OpenCV, Embedded Device
2. Survey on Variational Scene Flow Estimation
The aim of the paper is to study various variational methods for scene flow estimation and classify them based on the camera setting and the energy formulation. The paper further focuses on to asses the optimization algorithms used in the literature and deduce a logical comparison in terms of deviation from the ground truth of different data sets.
Keywords — Scene flow, Optical flow, Disparity, dense flow, Variational Method, Optimization, Successive over smoothing SOR, Coarse-grain smoothing
3. Behaviour based Control Strategy for Double Ackermann Steering Control of Autonomous Tandem Road Rollers
The projects explores the trajectory tracking application for road rollers during edge compaction task. The roller receives path information from the paver through a remote interface. The algorithm extracts the edge spline of the path and then is offset to create a trajectory for the rollers to maneuvre. The behaviour based control strategy allows the rollers to switch to crab steering mode of the "Edge Compaction mode is activated.
Keywords — Multi-Robot System (MRS), Navigation, Trajectory Creation, Trajectory Tracking, Autonomous Vehicle, Sensor Fusion, Behaviour Fusion, Remote Interface, Autonomous Vehicle Architecture
Tools — C++, Finroc (ROS inspired platform), Fingui, Unreal Engine
4. Multi-Robot Formation Control using Graph Theory
The project is an attempt to achieve the decentralized multi-robot formation control with the help of graph theory. In particular, we have studied different characteristics of multi-robot systems and formation control strategies associated with them. The project consists of of three experiments involving the square formation control of four unicycle robots using decentralized coordination.
Keywords — Multi-Robot System (MRS), Graph Theory, Decentralized System, Consensus Bias.
Tools — MATLAB, Simulink
