Projects

FLUENT Robotics Lab

Working on developing predictive model algorithms than can infer human helpfulness towards robots in social spaces by incorporating individual factors of latent helpfulness and the contextual factors of human busyness and robot frequency of asking. with the goal of enabling safe and intelligent robot-human interactions.

Contributed to the system level software development of the perception and navigation system for the two wheeled diffrential drive Stretch RE2 robot equipped with Intel realsense D435i camera, IMU, Wheel encoder and a Cartisean arm manipulator.

Responsibilities included incorporating key perception capabilities such as Human Action Recognition, Human Pose estimation, Panoptic segmentation and object detection, as well as Visual SLAM system using RTABmap and Move-Base for motion planning as part of the indoor navigation stack.

Evaluated multiple Visual SLAM (RTABmap) and Visual-Inertial SLAM (Kimera, VINS, ORB-SLAM) systems, focusing on precision in pose estimation and computational efficiency. Performance metrics, including CPU usage, were benchmarked against a Ground Motion Capture System (Opti-Track) for accuracy verification.

Robot Perception and Navigation
Visual/Visual-inertial SLAM (RTABmap, VINS, Kimera, ORB-SLAM)
Motion Planning(Move base)
Git
C++
Python
Pytorch
CUDA
Bash/Shell scripting
Deep Learning
ROS
ROS2
Docker
Cmake
Opti Tracker and Motive
System Integration
Human Robot Interaction
Full stack software development
Human Action recognition (PyslowFast)
Human Pose estimation (OpenPose)
Object detection and segmentation (Detectron2)

Biologically Inspired Robotics and Dynamical Systems Lab

Devoloped a perception and SLAM stack to enable autonomous naviagtion of heaxpedal robot for picking dandleions

Developed a perception pipeline using Intel realsense L515 camera to detect, track and create a map of dandelions.

Incorporated computer vision methods and clusterning algorithms to detect and produce a semantic mask of round yellow dandelions while tracking them consistently across time.

Engineered an autonomous environmental mapping system using an intel realsense L515 camera, leveraging SLAM techniques to enable precise localization and exploration. Key features included real-time odometery estimation, AprilTag-based loop closure detection, and geometric consistency verification.

Devoloping safe motion planning algorithms for the hexapedal robot using affine geometric heat flow and autonomous legged hill ascent

Agricultural Robotics

3D-Lidar SLAM

Motion Planning

Graph-Based SLAM

GTSAM

Google Ceres Solver

Python

C++

Git

Cmake

Numpy

Clusterning Algorithms (DBSCAN)

System intergration

April tag based Loop closure system

ROS

Machine Learning

Non-linear least squares optimization

Distributed Aerospace Systems and Control Laboratory

Worked on the implementation of self-triggered control using Control Barrier Functions to enhance safety and efficiency in critical systems.

Explored CBF fundamentals, including safe sets, barrier certificates, and stability conditions, and demonstrated its role in ensuring system states stay within predefined safe regions.

Implemented various design techniques, including optimization-based methods and quadratic programming, and explored modeling and robust control techniques for ensuring robustness against parameter variations in CBF-based control.

Applied CBFs to a double integrator system to showcase their real-world use in safety-critical control, and Developed a CBF-based control law effectively constraining system states within predefined safe limits.

Designed a controller addressing limitations of traditional periodic controllers. and intoduced the concept of a "safe period" for implementing Zero-Order Hold (ZOH) control, ensuring safety guarantees.

Safety Critical Control systems

Adaptive Control

Control Barrier Functions

Optimization

Quadratic Programming

MATLAB

Syrma SGS (Internship)

Designed a concept for a 6-DOF arm-robot ultrasonic welding system aimed at enhancing high-volume production yield in a semi-automated RFID-tag production line.

Created a simulation using GAZEBO and ROS, illustrating an arm-robot performing welding tasks.

Enhanced the project by implementing 'teach and repeat' waypoint navigation to demonstrate assembly line actions.

Compared manual operation processes with the robot's automation over a specific time period.

Test results revealed a substantial potential increase in production yield. Achieved yields of up to 98.8%, highlighting the significant impact of automation in industrial settings.

Assembly Line Automation

Robot simulation

ROS

Gazebo

Python

Trajectory Optimization

Experience

FLUENT Robotics Lab

Biologically Inspired Robotics and Dynamical Systems Lab

Distributed Aerospace Systems and Control Laboratory

Syrma SGS (Internship)