We consider the problem of building an effective quadrotor flight coordination model using natural language inputs. With recent advances in 3D rendering techniques such as Gaussian Splatting, training quadrotor flight vehicles in virtual environments such as physics engines has gained prevalence. Through randomization of task relevant objects and the surrounding virtual environment, previous research work has shown training in these virtual training environments transfers well to real world drone tasks.

I’m excited to join SuperUROP to continue building my UROP experience as an aspiring future ML researcher. SuperUROP’s emphasis on writing and diligent research reports will help develop important communication skills I’ll need in my future work. I’m excited to continue learning about 3D rendering while also crafting a final report to show my findings.