UROP Openings

Have a UROP opening you would like to submit?

Please fill out the form.

Submit your UROP opening

Reinforcement Learning Agents for human navigation models




4: Architecture

Faculty Supervisor:

Takehiko Nagakura

Faculty email:


Apply by:


Paloma Francisca Gonzalez Rojas: palomagr@mit.edu

Project Description

We are using human trajectory data from drone videos from Machu Picchu to feed to a Reinforcement Learning Model in Unity3D. We need data processing which involves extracting the trajectories from aerial videos with OpenCV tools and preparing it for the model. First the videos will be classified by valid trajectories, and also an algorithm needs to be developed to track all the visitors, even without valid trajectories, when they start moving. Then the trajectories need to be extracted, using x,y, coordinates and time stamp. The data then is visualized to inspect it. Indicators will be extracted from the data, like for example if a person stays static in the same position for a certain amount of time. Then the data needs to be discretized in order to be able to feed it to the Reinforcement model in Unity3D, following a model provided by me. It is a parallel goal of the project to co-create a routine to automate the trajectory analysis process.


Python Medium Level for use with OpenCV. Experience with OpenCv and other Computer Vision algorithms is desirable. Video editing basic skills and documentation skills.