Research Project Title:
RL Applied to Dynamic Scene Graphs: Model Selection and Optimization
abstract:RL can be applied to navigation by guiding an agent through an environment to a goal as quickly as possible. The agent is then assigned a score based on its success in reaching the goal state. In this project, the environment is represented as a Dynamic Scene Graph (DSG), in which nodes represent entities in the scene, and edges represent relations among nodes. We apply Graph CNNs to more accurately map the scene graph and agent position to a minimum number of actions that enable the agent to earn maximum reward, and use RL to train the agent to take a sequence of actions that will maximize reward. I will be working on model optimizations for Graph CNNs to improve performance of the RL pipeline.
Participating in SuperUROP will help me extend knowledge I have gained in classes like 6.141 and my previous UROPs, and enable me to explore different fields in robotics. I hope to learn what goes into building scalable and reliable robotic systems, from a solid foundation in research, to testing and deploying products in the field. I will also be able to form meaningful connections with professionals and researchers in industry and academia.