One of the main challenges with the complex dynamic systems found in legged robotics, especially when it comes to robustness is to identify and control situations where the sensors become unreliable. Such is the case with slippage, falling, or highly dynamic movements such as leaps. Reinforcement Learning offers an avenue for the robot controller to be learned to deal with these type of situations in close to an optimal way, and a similar approach can be used to identify these situations without time consuming tuning on the sensor fusion methods through hundreds of parameters. My work will focus on developing the algorithms for a novel bipedal humanoid robot to identify highly dynamic and unreliable situations in a consistent way, through the aid of an event camera.

SuperUROP exposes me to advance research work in the field of robotics. I aim to make good contributions to the lab through the program while also going deeper into the topics of novel control and estimation. The application of complex dynamic robotics is a fascinating one, and by the end of the year I hope to have made a publishable contribution to the field.