Modified the existing centerpoint detector to improve its performance on the 3D object (car, cyclist and human) detection. I tested the following add-ons.
Deformable Convolutions - Replaced the conventional convolution layers in the detector’s head with deformable convolutions. I conducted two sets of experiments: one applying deformable convolutions in the shared head, and another using them in both shared head and separate heads.
Data Fusion - Generated a BEV representation from camera RGB data and integrated it with the LiDAR BEV.
Attention Head - Included sliding window Self Attention in the detector’s head to enhance the modeling of long range geometric relationships.
RRT* Algorithm
Designed a motion planning system for warehouse drones using RRT*, integrating path generation, smoothing, and controller implementation for efficient navigation.
Wrote RRT, RRT* and informed RRT* algorithm from scratch and compared their performance for path-planning of drones in a warehouse environment.
Crane Operations Simulation - Haptics and Human Robot Control
How does visual and haptic feedback affects performance in cranes task involving placing package on target under varying wind condition.
Cart Pendulum Model
Modeled a 2 DoF cart and pendulum system with state space representation. MATLAB code takes mass, damping and stiffness coefficient and initial conditions of the cart and pendulum and plots their position and velocity curves along with animation.
I am a research candidate at Thapar Institute of Engineering and Technology, Patiala, working on Condition Monitoring of Rotary Systems under the guidance of Dr. Ashish Purohit and Dr. Jaskaran Singh.
I worked on developing the technique to classify the severity of mass imbalance in the rotary system attached to the electric motor. I used motor current signature analysis (MCSA) and machine learning algorithms to detect the mass imbalance and classify its severity. My work involved the development of experimental rigs, data acquisition, data analysis, and training and testing various classification machine learning models. This work is currently under review for publication in a Journal.
I participated in the International Rover Challenge (IRC-2023), leading the rover team's mechanical department under Dr. Ashish Singla's mentorship. Our team stood 6th out of 30 teams internationally.
I worked on the design and fabrication of the sub-systems of the rover, including chassis, rocker-bogie suspension, soil collection and analysis mechanism, suspension system, and robotic arm. My task involved performing design calculations, making CAD and fabrication drawings, and stress analysis of critical components.
I completed a one week training at Data Patterns (India) Ltd., Chennai, as part of the ThaparSat student satellite program. The training covered microsatellite subsystems, ground station systems, ADCS, satellite communications, and subsystem testing, including antenna deployment and RF checks.
I later contributed to the ThaparSat project by integrating software defined radio with the antenna system to receive NOAA weather images and by assisting in the mechanical design of the antenna base structure.
I co-founded a startup working towards the development of educational robotic kits. The main idea of our startup is to develop robotic kits that are fun to play with and have educational value.
We developed "Wheelo 1.0", an educational mobile robot kit, along with a dedicated course on mobile robots. The course covers essential topics like robot kinematics, Arduino programming, line following, etc., and teaches to implement these concepts on the physical hardware using the Wheelo 1.0 robot. "Wheelo 1.0" kit comes with an easy-to-assemble robot frame, custom PCB, microcontroller, instruction manual, and custom sheets to drive the robot.
We have conducted workshops in schools and colleges. Currently, orders can be placed by directly writing to bottedtechnologies@gmail.com. We also plan to list "Wheelo 1.0" kits on E-commerce platforms.