sherin joseph roy
I build software that drives cars. Currently developing a vision-first, Phase-2 autonomous prototype on a Tata EV platform. My focus is on End-to-End Behavioral Cloning—bypassing t... Show moreI build software that drives cars. Currently developing a vision-first, Phase-2 autonomous prototype on a Tata EV platform. My focus is on End-to-End Behavioral Cloning—bypassing traditional modular robotics pipelines by training lightweight convolutional neural networks (like PilotNet) to map raw camera pixels directly to physical steering and braking commands.
I handle the full AV stack: from hardware integration (NVIDIA Jetson, GMSL cameras, RTK GPS) to CAN bus reverse-engineering, data pipeline curation, and deep learning inference on edge compute.
Core Stack: Python, C++, ROS 2, PyTorch, OpenCV, CAN-FD. Show less
Trying to build a self-driving car from Bangalore, India. Specifically, building a Phase-2 Drive-by-Wire prototype on an EV platform using end-to-end behavioral cloning and NVIDIA Jetson edge compute.
Deployed a highly optimized, 250k-parameter PyTorch behavioral cloning model (PilotNet) onto an NVIDIA Jetson Orin for real-time edge inference.
My self-driving AI’s absolute hardest edge-cases aren't rain or unmarked lanes—they are Bangalore auto-rickshaws cutting across three lanes of traffic.
In web dev, a bug crashes a browser. In autonomous vehicle dev, a bug crashes tons of metal.