Ali Ekin G
- Research Program Mentor
MS candidate at ETH Zurich
Robotics, deep learning, control systems
BioI graduated from Princeton University's Mechanical and Aerospace Engineering department with minors in Computer Science, Robotics and Machine Learning. I am currently pursuing a master's degree in robotics at ETH Zurich. I have a strong theoretical and practical background in both the hardware and the software side of AI & Robotics. My current research addresses the safety and robustness challenges that are inherent in deep learning-based robot planning and control methods. In the past, I have interned as a software engineer in Ford's self-driving truck R&D team. Since my high-school years, I have done multiple independent and group projects in AI & Robotics, and I would love to mentor younger bright students in similar projects.
Deep reinforcement learning based planning for robot navigation
In reinforcement learning (RL), the agent learns a policy to maximize a cumulative reward by trial and error. Deep reinforcement learning combines RL with deep learning to learn complex robotic skills by uncovering the lower-dimensional latent space representations (via neural networks) that are relevant to the planning task. I have worked on deep RL-based systems in the past and I am interested to apply it to different application domains (ground or aerial vehicle navigation, robotic arm manipulation, etc.) as well as address various robustness and generalization challenges.
AI Playing GeoGuessr
GeoGuessr is an online game where users are presented with Google street view images and are asked to predict the location as close as possible. I am interested to work on a project where we develop an AI that can achieve (at least) human-level performance on GeoGuessr. This would be a comprehensive ML project where we would start by creating a dataset of Google street view images labeled with longitude/latitude. Then we can train a CNN or RNN based model that can predict a geolocation given the Google street view images. Later, we can also try to improve the performance of this model using a reinforcement learning approach. * GeoGuessr: https://www.geoguessr.com/ * An example of NN-based geolocation estimation: Paper: https://paperswithcode.com/paper/geolocation-estimation-of-photos-using-a Compete against the AI they built: https://labs.tib.eu/geoestimation/