Ashwin Vinoo - Robotics

ROBOTICS: A collection of projects created along with former classmates and/or in collaboration with the faculty of SCTCE labs, Sree Chitra Thirunal College of Engineering and Share Lab, CORIS, OSU. Most of these projects have been created to satisfy personal ambition and curiosity, some as a part of the coursework at OSU and a few others were to satisfy the needs of our clients. Other projects undertaken while working at Mu-Sigma are not mentioned in detail in this section to maintain client confidentiality.

1. RoboArtist

Created a robotic arm that can sketch portraits using a pen. Uses a Webcam to capture images, after which image processing and edge detection is performed via MATLAB. The pixels on the edges are mapped to servo angles and simultaneously fed to an arduino microcontroller that controls the arm. Currently, it takes around 30-40 minutes to sketch a detailed image of a person’s face. Links: RoboArtist PDF, RoboArtist PPT, Roboartist Code, RoboArtist Hackaday, RoboArtist YouTube

2. Autoencoders to enhance Multi-Agent Coordination in a Tightly-Coupled Domains

Deep reinforcement learning has been shown to address several multi-agent coordination tasks. However, with the increase in the number of agents as well as in tightly-coupled settings, learning optimal joint policies becomes challenging. In this paper, we address these two problems in a partially-observable rover domain environment using Deep Deterministic Policy Gradient (DDPG) and autoencoders. Links: Project Paper, Video 1: Original State Space of size 40, Video 2: Compressed State Space of size 10, Video 3: Compressed State Space of size 5

3. Efficient Robot Motion Replanning in Dynamically Changing Environments

Changes may occur in static maps due to introduction of obstacles, other agents or some other changes in the environment that render the path unusable. As there is a need to address efficient re-planning in dynamic scenarios, in this paper, we investigate an algorithm that can store a road-map of the environment and use it to efficiently recompute a path to the destination in the advent of such environmental changes. Finally, we compare this algorithm with other standard path planning algorithms like A* and RRT, to study the performance increase in computation time and path length obtained by leveraging the same. Links: Dynamic Path Planning Paper, Python Code

4. Automated Flame Extinguishing Robot

Created a robot that uses image processing and a IR sensor based heat detection panel to identify flames and extinguish them using an onboard pump and sprayer nozzle. The images are captured through an IP camera, processed in Matlab on a remote desktop and finally control instruction are fed to an onboard Arduino microcontroller via a Bluetooth dongle. This robot can also be controlled manually using an Xbox controller or through voice commands as well. Links: Flame Manager PDF, Flame Manager Code, Flame Manager 2 Code

5. Eye Controlled Bot

Constructed a two wheeled robot to simulate a wheelchair. Implemented a novel eye tracking algorithm in Matlab to identify the pupil and estimate the direction in which the user is looking. If the user looks left, the bot rotates anticlockwise and if he looks right, it will rotate clockwise. Similarly, looking up or down causes the bot to move forward and backward respectively. Later this project was implemented in an actual wheelchair by welding geared motors directly to the wheels. Links: Eye Controlled Bot PDF, Eye Controlled Bot Code

6. 4G Network Quadcopter

Created a Quadcopter that can be controlled through the 4G network. Uses an onboard camera to stream video, so that the user can control the quadcopter from any location. A large 5500mAh battery extends the flight time to around 20 minutes. Currently, flight stabilization code is developed on a raspberry Pi while an Arduino microcontroller handles communication to the several ESC’s, gyroscope and the accelerometer. A lightweight carbon fiber frame with propeller guards has been used to minimize damages inflicted during crashes. Links: Eagle Eye PDF, Eagle Eye Code

7. Coin Picker

Built a robot that can identify coins based on their color and shape and pick up the ones specified via voice commands. As the robot moves over the coins, their images are continuously captured using a camera attached to a raspberry Pi and then processed in python. Once the robot has identified a required coin, a robotic arm constructed using two stepper motors and bump switches collects the coin using an electromagnet. For all control purposes an Arduino microcontroller has been used. Links: Coin Picker PDF, Coin Picker Code

8. Corhnole Bag Tossing Robot

Designed a robot in Solidworks that can toss a half pound bean bag accurately into a cornhole table 3.4m away. The robot uses a catapult mechanism with a specially constructed rubber based spring to launch the bags. The spring is extended with the help a high torque wiper motor. A servo based loading mechanism was also designed to insert bean bags into the catapult head one by one. The robot can be navigated remotely using an online portal via camera feed. The physical implementation was carried out on the Pioneer PD3X platform. Links: Advanced Robotics Team 13 Website

9. Child Reward Module

Designed a reward stack in Solidworks that can offer a wide variety of light, sound and bubble rewards to infants to facilitate attention and learning. The design includes that of the user interface through which the user can scroll through several menus and change numerous settings. The top module was a custom designed bubble production mechanism that could launch bubbles approximately 7 inches into the air. The entire structure was 3D printed and could be powered by both a 12V jack or internal NiMh batteries. Links: Arduino Code, Solidworks Files

10. Disabled Infant Powered Mobility Training (Ongoing)

The child reward module was interfaced with a Turtlebot 2 robot and controlled remotely using a dualshock 4 controller. This was done with the intention that a disabled infant placed in a powered cart nearby would be more motivated to learn to use the cart, if light, sound and bubble rewards were provided when they were able to get the cart to move forward. The carts used for these studies were also hacked to interface with a datalogging unit to record the exact timing of when the infant pressed the button to move forward. Links: Python Code, Solidworks Files

11. Solidworks Vehicle Safety Device

Created two versions of a 3D printable plastic enclosure in Solidworks that would be used in constructing a vehicle safety device. The first version was designed to fit in a car’s cup holder tray, while the second could be directly plugged into the vehicle’s 12V power socket. This project was initiated to assist a local emergency doctor in his ambition to use technology to reduce mortality from accidents, worldwide. Links: Solidworks Vehicle Safety Device PDF, Solidworks Vehicle Safety Device Models

12. Automated Cradle

In order to assist Kerala government manage the ~144 child welfare centers around the state, our team was asked to design and prototype an automated cradle based system with CCTV based remote monitoring services, that could be used to place abandoned babies in. My task was to help in the initial prototype design, manage the CCTV system, prepare documentation to be shared with the concerned authorities and provide consultation on the approach to be taken. Links: Automated Cradle Image and Newspaper Articles, Solidworks Files

13. Large Scale Generative Mark Maker

Helped Aidan Phillips with his V plotter based project called "Large Scale Generative Mark Maker". My main contribution was to design and develop a new pen holder for the V plotter that was lighter and more robust. Then, I designed and fabricated a load cell based touch interface device that was used for live user interaction at an art exhibition where the project was displayed. Additionally, I helped him brainstorm new pen lifting mechanisms and develop an algorithm to regulate plotting acceleration and deceleration speeds to improve drawing quality. Links: Team GitHub, Solidworks Files

14. Autonomous Building Floor Activity Profiling System

Used Sequential localization and mapping (SLAM) in ROS to recreate the floor map of the robotics hall at OSU. Every hour, a PR2 robot with an on-board computer would navigate along waypoints in the map using Navstack. At each waypoint, we used a deep learning based face detection algorithm to identify faces at each location. The aim was to create a heatmap in Tableau that allowed the user to view floor occupancy levels at different hours of the day so as to effectively allocate hardware resources throughout the lab. Links: Project Paper

15. Child ELectric Car Hacking

As a part of my part-time work at Oregon State University, I was involved in designing the electrical system and microcontroller software that allowed electric carts for infants to record the exact time of movement on an onboard SD card. An L298N based motor driver board was used to control the geared motor. Additionally, an LED based error warning system and potentiometer based speed regulator was also installed behind each car. 12 such cars have been developed so far. Links: Arduino Code

16. Package picking and segregation using robotic manipulator

Created a simulation of an industrial Kuka 14 robotic arm that used a kinect depth camera to track and identify objects of different shapes, post which they were picked up by the arm and sorted based on their color. Multiple conveyor belts were used to move and sort the objects and parallel processing based control was implemented in python such that conveyor belts and the robotic arm could operate simultaneously. Links: Code and Simulation