Toronto, Ontario, Canada
Develop robotic manipulation techniques using deep reinforcement learning, integrating vision-based algorithms for real-time perception and control. Work on humanoid components to enhance dexterity and human-like adaptability in robotic systems.
Supported strategic decision-making by conducting in-depth analyses on engineering projects and operational processes from the Office of the CEO Drove collaboration within Magna’s 8 businesses to identify and implement efficiency improvements, promote innovation, and inform executive-level decisions to ensure long-term success. Led and developed new software and electronic prototypes; arranged a partnership with a leading auto-semiconductor company
Build motion retargeting pipelines that transfer 8000+ high-fidelity human tennis demonstrations to humanoid robots, forming the foundation for policy learning in dynamic, multi-contact athletic environments. Develop reinforcement learning and behavior cloning frameworks to unlock athletic capabilities in humanoid robots, leveraging human motion capture priors to achieve human-like coordination, balance, and agility.
Develop robust self-driving vehicles in Prof. Daniela Rus's lab using the MIT-developed car simulation engine VISTA Test models on a Lexus car modified with sensors, cameras, and a LiDAR system to understand real-world behavior Applied High Order Control Barrier functions, building upon BarrierNets applied to a neural network-based controller Incorporated techniques from control theory, integrating vision algorithms developed by Meta AI such as DinoV2
Instructed high schoolers on AI and Machine learning in the CogWorks course through project-based learning Managed Github organization for the course compromising of over 250+ students, instructors, and faculty Led a project on a patent-pending AI song mixer using Spotify’s dataset of over 100,000+ songs and YouTube analytics
Conducted research on robotics and reinforcement learning extensively using PyTorch, bridging the gap between simulation and real world robot behavior Trained over 80 million+ simulations of various robots in environments such as Nvidia Gym and Mujoco Implemented Proximal Policy Optimization (PPO) algorithm developed by OpenAI to enhance robot locomotion Gained expertise in cloud computing and parallel computing utilizing MIT’s cloud supercomputer SuperCloud
Analyzed the city’s data on business vacancies in Albany, inspected over 200,000+ vacancies and crime rates, developed visualization map software using Google API, presented findings to mayor’s office for city development