UROP Openings

Have a UROP opening you would like to submit?

Please fill out the form.

Submit your UROP opening

Relational Model-Based Reinforcement Learning and Planning with Applications to Robotics (Remote)


Term:

Spring

Department:

6: Electrical Engineering and Computer Science

Faculty Supervisor:

Leslie Kaelbling

Faculty email:

lpk@csail.mit.edu

Apply by:

11/20/20

Contact:

Tom Silver: tslvr@mit.edu

Project Description

We would like to be able to drop a robot into an unfamiliar simulated kitchen, have it playfully explore its surroundings, have it watch demonstrations of cooking and cleaning, and have it practice until it becomes an expert. Some subproblems to this end include: 1. Learning symbolic abstractions of the low-level environment 2. Learning relational, task-level transition models 3. Learning to plan quickly with those transition models We are looking for UROPs who are interested in pursuing projects in these directions, contributing to our long-term goal of a unified agent architecture for planning, learning, and playing. See https://web.mit.edu/tslvr/www/ for some of our previous and current steps in this direction.

Pre-requisites

Requirements: - Proficiency with Python - Machine learning (6.036 or equivalent) - Curiosity