Tigran Gasparian

Everyday Robots is a project from the Moonshot Factory (formerly Google X), an experimental project division of Google with the goal of creating general-purpose robots. I was part of this project from the phase when it was part of the Moonshot Factory until it got spun out into a separate company under the Alphabet umbrella.

We leveraged machine learning in many aspects to develop the robot and enable general-purpose capabilities. To develop and test the code running on the robot, as well as to train machine learning models, we used a simulator. This simulator needs to be accurate enough so that tasks performed and trained in a simulation can be applied to the real world with as few modifications and refinements as possible.

In this simulation team, I worked as a graphics programmer, where I redesigned and rewrote the entire renderer used to simulate camera sensors, enabling around 10x performance improvement when running the simulator in Google's internal cloud, Borg. It also improved the graphical fidelity of the images and enabled the simulation of additional modalities such as IR and UV and sensors like fisheye cameras.