Gautham Narayan Narasimhan

Hello! I am currently a Research Assistant at the Robotics Institute (RI) at Carnegie Mellon University. Previously I completed my masters thesis with Prof. David Held at CMU.

I study machine learning algorithms that enable robots to perceive and manipulate general purpose objects. Currently I'm working on model based reinforcement learning for robotic pouring tasks.

Contact: gauthamnarayn (at) gmail.com

Resume / GitHub / Google Scholar / LinkedIn

News

June 2022 : Carnegie Mellon SCS wrote an article on our work in Robotic Pouring - [Link]

Jan 2022 : Self-supervised Transparent Liquid Segmentation for Robotic Pouring accepted to ICRA 2022!

Oct 2020 : ROLL: Visual Self-Supervised Reinforcement Learning with Object Reasoning accepted to CoRL 2020!

Invited Talks

July 2022 : Intel Embodied AI Lab - Transparent Liquid Image Segmentation For Robotic Pouring [Slides]

Research and Publications

	Learning Material Properties using a Differentiable Simulator for Liquid/Granular manipulation Gautham Narayan, Xingyu Lin, David Held In progress Currently training a Material Point Method based differentiable simulator to learn material properties. Cross Entropy Method has shown good performance for trajectory optimization once the material properties have been learnt. More details coming soon ...
	Self-supervised Transparent Liquid Segmentation for Robotic Pouring Gautham Narayan, Kai Zhang, Ben Eisner, Xingyu Lin, David Held ICRA 2022 and abridged at NeurIPS 2021 Deep Generative Models Workshop A novel segmentation pipeline that can segment transparent liquids such as water from a static, RGB image without requiring any manual annotations. We show that this system can run in real-time and aid in tasks such as robotic pouring. [Paper] [Website]
	ROLL: Visual Self-Supervised Reinforcement Learning with Object Reasoning Yufei Wang, Gautham Narayan, Xingyu Lin, Brian Okorn, David Held * denotes equal contribution Conference on Robot Learning, CoRL 2020 Unknown object segmentation to learn a visual representation that can reason about occlusions. Our method achieves state of the art on object manipulation benchmarking tasks. [Paper] [Website]
	Segmentation for learning image based goal conditioned policies Gautham Narayan, David Held Master's Thesis - Carnegie Mellon University, 2020
	Experimental Droplet Spatter Analysis Using Least Squares Approximation Gautham Narayan, Bill Eddy Internal Report - NIST Center of Excellence in Forensic Science, 2020
	Effect of winglets induced tip vortex structure on the performance of subsonic wings Gautham Narayan, Bibin John Elsevier - Aerospace Science and Technology, 2016 Design optimization for subsonic winglets using computational fluid dynamics. [Paper]

Source