Toru Lin
MIT EECS | Aptiv Undergraduate Research and Innovation Scholar
Learning Particle Dynamics in Partially Observable Scenes
2018–2019
EECS
- Robotics
Antonio Torralba
The goal of this project is to build a model that is capable of generating particle representation and predicting the dynamics of partially observable physical scenes. This is a step forward from Dynamic Particle Interaction Networks (DPI-Net), a particle-based simulator that can only learn object dynamics and make predictions based on fully observable scenes. Upon successful implementation of the new model, we will be able to create robots that can quickly adapt to new environments with unknown dynamics and accomplish various real-world control tasks using less computation.
“I am excited about SuperUROP because it will allow me to gain a yearlong experience in academic research. This project interests me a lot because it lies in the intersection of robotics and artificial intelligence. Having worked as a software engineer, I look forward to utilizing my engineering skills and diving deeper into the optimization of robotic control with the aid of machine learning.”