For my Master's thesis research, I worked on applying Reinforcement Learning models to real-time Dynamic Scene Graphs (DSGs) on hardware. This builds up on my senior thesis research where I worked on adapting the RL stack to accept DSGs provided by Hydra (a real-time Spatial Perception Engine) as input, and evaluated in simulation. My research aimed to test the accuracy of the RL models while running them on DSGs produced by a real-time SPIN on hardware. I also investigated how to reduce both the computational demands of generating DSGs and the latency of inference for this pipeline such that inference can run in real time. 

 

Styled With

MIT SparkLab

You may also like