Human Robot Interaction using Leap Motion Controller in Virtual Reality
Implemented Human Robot Interaction (HRI) using a leap motion controller in Virtual Reality(VR). The primary objective was to make a robotic arm mirror the movements of a human in a factory setting.
The work included a haptic feedback as measure to "feel" objects in the virtual world and a "menu on hand" as can be seen in the VR world.
Developed the program for "holding objects" using the leap motion controller in Unity using C#
Tested the program on Oculus Rift
Stabilization of load blocks in an Overhead Crane using Linear Quadratic Gaussian Controller
As a final project for the course ENPM808Q, we designed a LQG controller for a system which was very similar to a Overhead Crane (but with two load blocks).
We designed the LQG Controller for stabilizing the "load blocks" in the system. By controlling the "trolley frame" while providing a constant reference input signal, we successfully managed to stabilize the block as well as guide the load block to the desired location on the "girder".
We simulated the same in SimMechanics with very good results.
We also developed a Linear Quadratic Regulator controller for this system after linearizing it about its equilibrium point.
Design and simulation of Revolute Input Delta Robot
I designed a revolute input Delta Robot with the following features:
It is a 3 Degrees-Of-Freedom robot with three identical legs in parallel between the fixed “base” at the top and a similar end-effector platform at the bottom.
The revolute joints at the “base” are actuated via base-fixed rotational actuators. The zero angle defined as when the actuated link is in the horizontal plane.
The parallelogram 4-bar mechanisms of the three lower links ensure a motion which is purely translational in nature.
Designed the robot in a CAD software - Autodesk Inventor 2016 and imported the same to SimMechanics to control this robot model.
Calculated the Inverse Kinematics Solution for this robot and used it to control the model accurately.
I-ARM - A 4-Axis articulate serial manipulator
I-ARM is a servo based, vision guided, 4-axis, articulate, autonomous robotic serial manipulator, with memory, for pick and fetch operations.
This project was a good example of Reinforced machine learning and Visual servoing.
Key Responsibilities
Design and construction of the robot.
Embedded System design and programming in CodeVisionAVR.
Panoramic stitching - Generation of a Mosaic from two images
This project was done in three stages:
1) Keypoint Detection:
Used the standard VLFeat library for the SIFT functions to find the corresponding pixels in both images and found the best affine transformation for the two images.
2) Image registration
Implemented RANSAC to find the best affine transformation. Iterating the algorithm ensured an excellent affine transformation.
3) Image stitching
Used the affine from RANSAC to stitch the images. It produced excellent results.
Implementation of Rapidly exploring Random Tree (RRT) Path planning algorithm
Used MATLAB as the tool to implement RRT.
Implemented Goal biasing to improve the results of the path planning algorithm.
Utilized the MATLAB functions to generate an obstacle space and free space and showed that a path can exist from the initial point to the final point.
Considered two orientations of the mobile robot to justify a suitable global plan.
Implementation of Efros & Leung's algorithm for Texture Synthesis
Developed the code in MATLAB to generate the texture.
Used a 3x3 pixel seed for texture synthesis.
Checked the results with an increasing window size and observed the results.
Dynamic path planning for unmanned aerial vehicles (UAVs) - A survey
I classified the algorithms mentioned below into following four categories: Node based optimal algorithms, mathematical model based algorithms, bio-inspired algorithms, multi-fusion based algorithms.
Algorithms covered in this survey:
Combination of Focused Dynamic-A* and Bspline, Combination of A* and Fast Marching Method, Interfered fluid flow, Proportional navigation law, Model Predictive Control algorithm to create
a optimize control actions, Partially Observable Markov Decision Process(PODMP)
framework, Hybrid algorithm comprising of lyapunov vector field guidance and tangent vector field guidance, Combination of linear algorithm and modified dubin’s algorithm, Evolutionary algorithm, Improved Ant Colony algorithm, Combination of ADS-B and closed loop-RRT, Combination of improved RRT* and D*-Lite, Combination of voronoi’s diagram and dijkstra’s algorithm, Fusion of Intuitionistic Fusion Sets, RRT-Series, Reachability sets and Receding Horizon.
Design and simulation of a planetary surface rover
Designed and provided a complete CAD model of the rover. Included the mass budget of the rover weighing about a 100kg with a payload capacity of 35 kg. Simulated the CAD model in Autodesk Inventor.
Chose the components including the motors for steering and driving, GGPU and battery for maneuvering the rover over an obstacle size of 30cmX15cmX30cm. Included the power budget of the vehicle for a continuous operation of two hours.
Designed the grousers for the wheels and a rocker bogie suspension system considering the terramechanics of the lunar soil, and performed nelder-mead optimization on the geometric constraints to obtain the best wheel size and suspension system.