General requirements

• Each project can have up to two members on the team. The project grade will be calibrated by the size of the team.
• A project should include the following components:
  • It builds/applies a machine-learning model to an interaction problem. The interaction can be language interaction, behavior-only interaction, or a combination of both interactions.
  • Such interaction needs to involve robots (i.e., an embodied agent that can sense, reason, and take actions to alter its working environment). So, the project cannot be purely training a vision-language model for question-answering or a reinforcement learning agent for Atari games.
• The robots for the project can be simulated or real robots.
  • Here are some simulators you may consider using (*****We will introduce them in the tutorial)
    • *robosuite: https://robosuite.ai/
    • *RoboTHOR: https://ai2thor.allenai.org/robothor
    • CARLA: https://carla.org/
    • AirSim: https://microsoft.github.io/AirSim/
    • VirtualHome: http://virtual-home.org/
    • … you can choose any simulator that fits your project better, just explain that in the project proposal.
  • If you have access to a real robot and want to use it, you are very welcome to develop your project on a real robot. You will need to describe how you would like to involve the real robot (e.g. in training, fine-tuning, etc) for the project.

Project directions

We encourage you to explore ways to include learning in an interaction problem of your interest. Here are some directions you may consider:

• Improve an existing method: You can pick an existing work you’re interested in and improve it based on what you learn in class.
• Apply an AI/ML method to a new interaction problem: You will need to design a new task/interaction, understand the strengths or weaknesses of an AI/ML method, and apply the AI/ML method to the new task/interaction.
• Create a new method for a problem: You will come up with a new method for a specific interaction problem. This problem can be the one we discussed in class or something no prior work has explored before.
• Bring your own research problem: ****If you have an interaction problem in your research projects, you are very welcome to make it a course project. You will need to explain what will be new for this course instead of using your existing method or results.

Grading policy

The course project is 40% of your final grade. Here is the breakdown for the deliverables:

• Proposal: 5%. Due Sep 21.
• Project milestone: 5%. Due Oct 26.
• Demo video: 5%. Due Nov 27.
• Presentation: 10%. Due Nov 27 or 29.
• Final Report: 15%. Due Dec 6.

<aside> 💡 The deadline for submissions is 10pm ET on the specified date. Late submissions will incur a 10% deduction on the score for each late date, e.g., 10% for one day, 20% for two days, etc.

</aside>

Project proposal

The project proposal should be only one page. Your project proposal should have the following components.

• Team member
• (20%) Motivation/Background
  • Why this problem is important?
  • Give a motivating example/scenario
• (20%) Problem definition
  • Define the problem to be solved, e.g., input, output, and metrics/criteria of success