Medford, Massachusetts, United States
Hello! I’m passionate about all things robotics, from sketching initial designs to the thrill of watching a fully integrated system in action. With hands-on experience across the entire development cycle, I thrive on transforming complex robotic concepts into tangible, operational systems. This journey for me includes everything from modeling and simulation to manufacturing and fine-tuning the sensors and controls. Beyond the mechanics, I'm also deeply involved in the brain of robotics. I specialize in —3D computer vision techniques like object detection, semantic segmentation, and Visual Inertial Odometry (VIO), all the way to Simultaneous Localization and Mapping (SLAM). I’m adept at leveraging deep learning models to bring intelligence to edge computing devices, making robots smarter and more efficient. I love connecting with fellow like-minded tech enthusiasts and innovators who push the boundaries of what robots can do.
• Drafted and submitted an NSF NRI 2.0 proposal, resulting in a $1.5 million research grant for developing a self-balancing smart wheelchair. • Developed a CNN-based vision-proprioception fusion method for robust UGV terrain classification: o Used common sensors and CNN fusion models to identify terrain types by assessing the features like vibrations, slippages, and terrain appearances. o Achieved over 93% accuracy and high robustness under various lighting conditions and motion maneuvers on a Jackal UGV. o Published in IEEE RA-L and presented at the IEEE IROS conference in 2021. • Developed a self-supervised one-shot segmentation framework for robot navigation and anomaly detection: o Used a self-supervised methodology to eliminate the need for any dense manually labeled dataset for the training of this model. o Showed the framework's ability to identify small low-lying objects and yielded up to six times faster inference speed with comparable accuracy as compared to two state-of-the-art indoor semantic segmentation models. robots and robot runway patrol. • Developed a vision-guided shared motion control (SMC) system for the safeguarding of self-balancing ballbot wheelchairs: o Performed full system integration to ensure the low-level dynamic control module, HRI, and high-level decision module work in conjunction. o Designed noise isolation solutions to minimize electrical and EMF interference within the smart wheelchair system. o Developed a SMC system (similar to a driver-assistance) that provides features like controlled stop, collision avoidance, path keeping, etc. o Conducting a human subject study to assess whether the addition of SMC can reduce user proficiency requirements and collision risks. • Developed a dynamic simulation platform for validating ballbot drivetrain models and enabling control co-design. • Developing a Latent Diffusion Model (LDM)-based data-label generation method for generating rare or specific image datasets.
• Developed robot platforms, lab manuals, and homework assignments for the course “Autonomous Vehicle Systems” covering common practices in using cameras, Lidars, range finders and GPSs to achieve simple autonomous navigation and driving behaviors in ROS. • Taught and designed lab exercises/assignments and manuals for the course “Intro to Robotics” on robot kinematic and dynamic modeling, as well as implementing robot controllers in ROS and C++. • Led weekly lab sections for the course “Computer Control of Mechanical Systems” on creating circuit boards, implementing communication protocols like RS232, UART, TTL, I2C, SPI, and PWM, and programming embedded systems in C.
Designed, built, and programmed a mobile robot for the Mobility Unlimited Challenge, achieving a top 10 ranking in the final global competition.
• Developed a visual odometry codebase using SIFT features, FLANN and ICP matching to recover vehicle motions from dash camera videos (KITTI, openpilot). Created global bird’s-eye view road mosaics from videos recorded by front-facing dash cameras using IPM. • Developed software enabling users to label lane lines and other road features from a global bird’s-eye view image and project the labels back to individual front-facing images. It allows users to label repetitive features in bulk, which largely reduces manual labelling cost.
• Designed, built, programmed, and evaluated a miniature self-balancing home security robot prototype for clustered indoor environments. • Simulated the robot dynamic in Matlab Simulink and implemented a LQR controller for its remote motion control.