Sri Anumakonda

I'm an autonomous vehicle enthusiast building in the intersection between artifical intelligence and self-driving cars.

Current: Member @Masason Foundation, also am leveraging CUDA to train better semantic segmentation models while also learning math on the side.

Prev where I worked with Shell, the United Nations, and Instacart. Also am one of the youngest certified self-driving car engineers in the world, was featured on Udacity for the work I'm doing.

Email  /  Linkedin  /  Twitter  /  Github  /  Medium /  CV /  2021 Annual Letter

profile photo

My primary research interests are in the field of autonomous vehicles, more specifically towards computer vision for vision-based autonomy. You can find some of my projects in this space below. This is just an overview of the work I'm doing. For a full list, please check my Github.

clean-usnob CUDA-Optimized Semantic Segmentation for Autonomous Vehicles
February 2023 - Present

Leveraging CUDA + OpenCV + PyTorch's C++ API (LibTorch) using TorchScript to run semantic segmentation at the highest FPS possible

clean-usnob Creating Autonomous Vehicles with Deep-Q Networks
January 2022 - February 2022
RL Article / Tweet

Leveraged Reinforcement Learning algorithms to help train a self-driving agent to drive in a highway environment [using discrete controls].

clean-usnob End2End Learning for Lateral Control
November 2021 - January 2022
Research Proposal / Tweet

Used Deep Convolutional Networks to control self-driving steering, read more than 60 papers in the space, and met some really smart people from the + team!

clean-usnob DataGAN: Leveraging Synthetic Data for Self-Driving Vehicles
September 2021 - October 2021
Medium article / Github

Using DC-GANs to create synthetic data that can be used to train + validate the robustness of autonomous vehicles. End goal is to apply to extrapolate the data pool we have for adverse driving scenarios/any situations where limited data is available.

clean-usnob Convolutional-LSTMs for Lane Detection
July 2021 - August 2021
Project abstract / Github

Implemented this paper from scratch with just my knowledge of Python and stackoverflow. Focusing on managing spatio-temporal image data for successful training of lane detection. Lanes don't change every single frame, so why can't we take in a few input frames and be able to successfully predict lane lines? Note: Github repository is not up to date as I have been unable to locate some of the files as part of training but the ConvLSTM model and training loop is created (and formatted nicely!).

clean-usnob Building a self-driving car!
May 2021
Medium article / Certificate / Github

Used my knowledge throughout the Udacity Self-Driving Nanodegree [from perception and sensor fusion of LiDAR and radar sensors to polynomial trajectory generation] to create my capstone project; an autonomous vehicle capable of driving in simulation (using waypoitns).


Want them in your inbox? Add your email here.

Template from Jon Barron