About Me

Hi, I'm Parth Mahajan – a robotics engineer by profession, a painter at heart, and a beginner violinist exploring the world of music.

Professionally: I am passionate about building real-world solutions in motion planning, SLAM, and robotic perception, ensuring that innovation extends beyond simulations into practical, deployable systems. My personal motto is inspired by Feimen's: the fastest way to learn something is to teach it.

I plan to concentrate my career in Robotics on the Entrepreneurship/Innovation track while working on the software engineering team.

If our visions converge, let us connect and propel each other toward shared success!

Professional Timeline

Work Experience Projects

Please Refresh the page once in case the videos are not visible

Product AMR: DYNAMO

This video showcases the Autonomous Mobile Robot (AMR) product developed as part of a larger project at addverb, where I was responsible for designing, implementing, and testing the camera calibration and collision safety detection modules.
Additionally I was responsible for writing lower level drivers for lidar and ultrasonics Sensors
The camera calibration module ensures precise perception and environment mapping by refining sensor accuracy, while the collision safety detection system enhances real-time obstacle avoidance, ensuring safe and efficient navigation.
This contribution was integral to the AMR’s ability to operate autonomously and reliably in dynamic environments, reinforcing its practical deployment in industrial automation and smart mobility applications.

Please Refresh the page once in case the videos are not visible

Product Pick and Place Binpicking

This video showcases the Bin Picking system developed at Addverb, highlighting my key contributions in advancing its perception, motion planning, and control capabilities and was the major project of my two years at adverb.
I worked on feasible trajectory generation, ensuring smooth and collision-free motion planning for robotic arms.
Additionally, I was responsible for writing the SKU detection using perception algorithms to accurately identify the objects, as well as pick and drop pose estimation to optimize grasping efficiency.
To enhance system decision-making, I implemented and integrated a Behavior Tree framework, enabling modular, scalable, and adaptive task execution. This project significantly improved automation accuracy, efficiency, and adaptability in industrial bin picking applications.

Mehar Baba Project

Please Refresh the page once in case the videos are not visible

Indian Air Force Mehar Baba Project

Please Refresh the page once in case the videos are not visible

Decentralized Swarm Controller

This project stands as a testament to the countless hours of dedication, innovation, and perseverance by my team and companions at Unmanned Aerial Systems DTU. It represents the successful implementation and on-field demonstration of a Decentralized Swarm Controller, tested with 21 UAVs in the rugged deserts of Jaisalmer, Rajasthan. Conducted in preparation for the Indian Air Force Mehar Baba Prize, this project shows fully autonomous multi-UAV coordination system where each drone made real-time navigation decisions based on decentralized swarm intelligence. This project was developed with the purpose of large scale UAV search and rescue mission.

( To Know about the work , one can reach me through the contact Page)

This video presents a minimalistic yet effective approach to managing complex swarm behavior in Unmanned Aerial Vehicles (UAVs) through an extended decentralized, consensus-based control framework for multi-agent systems. The method ensures scalability and autonomy while addressing three key objectives: formation control for maintaining structured flight patterns, waypoint following for precise navigation, and static and dynamic obstacle avoidance to ensure collision-free movement. Designed to operate independently of the UAV count, this approach enables seamless coordination, real-time adaptability, and efficient mission execution across diverse environments.

( To Know about the work , one can reach me through the contact Page)

A Glimpse Of My Work

Please Refresh the page once in case the videos are not visible

Collision Free Diffusion With DDPM and Fast ESDF

This work demonstrates how differentiable guidance constraints can be effectively combined with diffusion-based motion planning to produce collision-free and dynamically feasible trajectories. By integrating ESDF-based guidance using nvidia's nvbloxx, the method achieves superior trajectory quality while maintaining computational efficiency.

This project holds promise for applications in autonomous navigation, robotic motion planning, and real-time trajectory optimization.

( To Know about the work , one can reach me through the contact Page)

Please Refresh the page once in case the videos are not visible

MOPSO : Modified Particle Swarm Optimization

This work demonstartes a decentralized method for multi-target search problem using a swarm of unmanned aerial vehicles with the information available from the onboard sensors. The proposed method deals with three main objectives: Time optimized multi-target search, optimized payload drops and inter-UAV collision avoidance which is independent of the number of UAVs.

( To Know about the work , one can reach me through the contact Page)

Please Refresh the page once in case the videos are not visible

Pure Pursuit Controller on Carla Simulation

This video showcases the implementation of a low-level Pure Pursuit controller within the CARLA Simulator as part of an autonomous self-driving stack. The Pure Pursuit algorithm, a geometric path-tracking method, enables the vehicle to follow a predefined trajectory by dynamically adjusting its steering angle based on a lookahead point. Integrated within the CARLA environment, this implementation ensures smooth and accurate path following, enhancing the self-driving system’s ability to navigate varied road conditions, turns, and dynamic scenarios. The controller plays a crucial role in maintaining trajectory fidelity, stability, and real-time responsiveness, making it a fundamental component of autonomous vehicle navigation.

( To Know about the work , one can reach me through the contact Page)

Please Refresh the page once in case the videos are not visible

Pure Pursuit Controller On Mahindra E2o Car

This video demonstrates the implementation of a low-level Pure Pursuit controller on a real hardware car, showcasing its effectiveness in trajectory tracking and autonomous navigation. The Pure Pursuit algorithm dynamically adjusts the vehicle’s steering based on a lookahead point, ensuring smooth and precise path-following in real-world conditions. This implementation validates the controller’s performance in handling varied road geometries, turns, and dynamic obstacles, bridging the gap between simulation-based testing and real-world deployment in autonomous driving systems.This was done at my time as an inntern at IIIT-Hyderabad at Robotics Research Center LAB.

( To Know about the work , one can reach me through the contact Page)

Please Refresh the page once in case the videos are not visible

Deep Q Learning Based Bang Bang Agent ( Training )

This video demonstrates the training and implementation of a Bang-Bang Reinforcement Learning (RL) controller that learns a Deep Q-Network (DQN) policy for obstacle avoidance using laser scan sensor data. Developed as part of my internship at Addverb, the model was trained in standard ROS simulation environments, where it learned to make discrete control decisions for safe and efficient navigation. By leveraging laser scan inputs, the controller dynamically adjusts movement to avoid obstacles while optimizing path efficiency. This project showcases the integration of RL-based decision-making in robotic navigation, paving the way for autonomous and adaptive motion planning in real-world applications.

Please Refresh the page once in case the videos are not visible

Testing Generalization Of Deep Q Learning Agent

This video demonstrates the training and implementation of a Bang-Bang Reinforcement Learning (RL) controller that learns a Deep Q-Network (DQN) policy for obstacle avoidance using laser scan sensor data. Developed as part of my internship at Addverb, the model was trained in standard ROS simulation environments, where it learned to make discrete control decisions for safe and efficient navigation. By leveraging laser scan inputs, the controller dynamically adjusts movement to avoid obstacles while optimizing path efficiency. This project showcases the integration of RL-based decision-making in robotic navigation, paving the way for autonomous and adaptive motion planning in real-world applications.