Sai Aneesh Suryadevara

ssuryadevara [at] ucsd.edu

I am a 1^st Year M.S. in ECE student at UC San Diego, specializing in Intelligent Systems, Robotics, and Control (ISRC). I completed my undergraduate from the Indian Institute of Technology Bombay, receiving an Honours degree in Mechanical Engineering and a Minor degree in Artificial Intelligence and Data Science.

Currently, I am a researcher at the Contextual Robotics Institue, working on Language-guided open-world mobile manipulation with a legged robot, advised by Prof. Xiaolong Wang . During my undergrad, I worked on my Bachelor's Thesis on Deep Reinforcement Learning for the Control of Soft Continuum robots advised by Prof. Abhishek Gupta and Prof. Shivaram Kalyanakrishnan. I also had the opportunity to intern at the University of Toronto, working with Prof. Lueder Kahrs at the Medical Computer Vision and Robotics (MEDCVR) lab for the summer of 2022.

Research Interests: Robot Learning, 3D Computer Vision, Embodied AI

Email / GitHub / Resume / Google Scholar / LinkedIn

News

Oct 2023: Started working as Graduate Researcher, advised by Prof. Xiaolong Wang
Sep 2023: Excited to start my graduate studies at UC San Diego!
Aug 2023: Our paper on RL for Surgical Robots has been accepted to IEEE-RAL!
May 2023: Graduated from IIT Bombay
May 2022: Will be joining the MEDCVR lab at the University of Toronto, as a Mitacs Research Intern

Publications

Learning Nonprehensile Dynamic Manipulation: Sim2real Vision-based Policy with a Surgical Robot

Radian Gondokaryono, Mustafa Haiderbhai, Sai Aneesh Suryadevara, Lueder Alexander Kahrs
IEEE Robotics and Automation Letters (RA-L) 2023

Paper | Webpage | Video | BibTeX

Research

Bachelor's Thesis: Control of Continuum Robots Using Deep Reinforcement Learning

Guides: Prof. Abhishek Gupta and Prof. Shivaram Kalyanakrishnan
[Code]

Implemented a model-free reinforcement learning approach to train control policies for trajectory tracking of a soft continuum robot arm. Developed a custom OpenAI Gym environment and integrated it with VEGA FEM C++ middleware library and ROS to simulate more realistic dynamics.

Decentralized Multi-Agent Patrolling using Q-Learning

Guides: Prof. Arpita Sinha and Prof. Leena Vachhani
[Code]

In this work, we wish to find an optimal patrolling strategy in a multi-agent setting with the constraint of minimum information sharing. Developed patrolling techniques and analyzed their performance using ROS, TraCI and SUMO simulator

Key Technical Projects

	Flipkart National Competition: Autonomous package delivery bots IITB Team Lead, National Semi-Finalists [Code] Developed a system of mobile bots capable of autonomous package sorting using ROS and OpenCV framework, tracking each bot’s pose through ArUco markers.
	e -Yantra Robotics Competition: Autonomous Delivery Drone System [Code] Simulated a working prototype of an autonomous drone in Gazebo for package delivery during Covid-19. Designed attitude and position (PID) controllers in ROS and implemented A* algorithm for path planning and obstacle avoidance.
	Image-to-Image Translation using CycleGAN and DiscoGAN GNR638: Deep Learning and Pattern Recognition for Computer Vision [Code] Implemented and compared the image generation capabilities of GANs and VAEs in PyTorch. Also investigated the performance of DiscoGAN and CycleGAN architectures for style transfer between Pansy and Tigerlily.
	Statistical Solvers using Graph Neural Networks IE643: Deep Learning Theory and Practice [Code] Worked on a paper implementation to understand Deep Graph Neural Networks as a new class of solvers for permutation-invariant optimization problems that can be trained without a training set of sample solutions
	Self Driving Car, University of Toronto MOOC, Coursera [Code] Built an environment perception stack, using a Semantic Segmentation neural network for lane estimation and object detection to alert the car about the position and category of obstacles