Marmik Thakkar

Robotics Software Developer

Senior Developer

• Work on strategic initiatives on technology selection, develop prototypes, research and development and process improvement
• Proactively identify and communicate risks & issues to management and create action plans to mitigate risks
• Continually improve our microservice architecture and performance
• Assist in the management, maintenance and automation Kubernets deployments
• Develop unit, integration tests, and validation tests
• Implementing telemetry services to enable distributed tracing for our microservices
• Frameworks & Tools: Nextjs, React, Nestjs, Graphql, Prisma, Docker, Jaeger, Opentracing, PostgresSQL, NoSQL, Redis, Nats, Rabbitmq, Bullmq, Jest, K6

Junior Developer

• Perform the smoke, load and stress tests to ensure service quality
• Develop and manage hybrid mobile apps and internal web apps
• Simulate end to end system to test all stakeholders' tasks
• Develop automated UI Testing to test multiple native apps
• Frameworks & Tools: React, React Naitve, Graphql, Appium, Jest, K6

Self Employed

Clinic Manager — Real-time web app that provides a suite of functionalities that makes it easy to manage a clinic. The system aids in managging the patients' records, the appointments, the doctors' schedules, prescriptions, manage the inventory, etc.

Transformation-Equivariant 3D Object Detection for Mobile Robots Utilizing Synthetic Point Cloud Data

This thesis proposes an approach to leveraging deep learning for the precise detection of objects in 3D point cloud. The core objective is to train a domain-specific deep learning model for manufacturing environments, utilizing the power of synthetic data generation, while also integrating it with the Robot Operating System (ROS2).

A framework utilizing a robotics simulation toolkit, such as Nvidia Isaac Sim, is implemented to generate labeled training data for supervised deep learning. The framework employs domain adaptation and randomization techniques, enabling models to generalize to real-world scenarios despite being trained on synthetic data. This approach aims to bridge the gap between simulation and real-world conditions.

Through multiple iterations of data generation and model training, the thesis explores the impact of employing synthetic data for training and its effect on object detection performance in real-world scenarios. The research offers a solution for training accurate point cloud object detection models for mobile robots, reducing the need for real-world data collection and optimizing the training process without the need of manual annotation. This framework holds promise for advancing intelligent systems in fields beyond mobile robotics, such as autonomous driving.

NMIMS Robotics Testbed - 1

Designed a payload, which is set to launch aboard the Indian Space Research Organization (ISRO)'s Polar Satellite Launch Vehicle (PSLV-C49). The payload is a testbed for a magnetostrictive material - Terfenol-D

Table Tennis playing Robotic Arm

An autonomous system that uses a stereo camera to map the ball trajectory 6 Axis robotic arm acts as an actuator to play table tennis with a human opponent.

Autonomous Underwater Vehicle

An underwater bot that is designed to do tasks that imitate ongoing research in Autonomous Underwater Systems such as acoustic navigation, shooting torpedoes, and dropping items.

Manual and Autonomous Bot

Manual and autonomous robots were developed to perform a series of tasks- delivery of a ball from a bot to another to throwing those balls through a hoop.

Home Automation using IoT

Manual and autonomous robots were developed to perform a series of tasks- delivery of a ball from a bot to another to throwing those balls through a hoop.