Logistics

Expectations

Generally speaking the project is expected to fall into one of the following options:

• Option A: Identify and tackle an open research problem.
• Option B: Application of sensorimotor learning to a specific problem of your interest. The problem can be from any domain.
• Option C: Or explore and implement advance ideas in sensorimotor learning that are inspired by the class content, but go beyond the material covered in the lecture (e.g., implementing ideas in some papers that were referenced in the class or that you found, but were not implemented in the homework).

The project must involve using at-least one algorithm (or its variation) or an idea covered in the class. The general hope is that your controller will involve “learning” in some aspect.

Grading

Projects will be “not” be graded based on whether you had good results, especially if you opt for options A and B. The outcome of research / applying learning algorithms to new domains is unpredictable and we embrace the reality. We will grade based on:

• Whether you were clearly able to identify the problem and motivate use of learning for control.
• The soundness of the methodological formulation.
• If your method did not produce expected results, how did you modify your initial method.
• The iterative process between method modification and the results.
• Your takeaways and conclusions from the project. It could be:
  • I got great results.
  • My results aren’t that good, but I learned this “...”

For option C, if you decide to implement an existing paper, we would like you replicate the results either in the domains presented in the paper or simpler version of the domains due to computational constraints.

What is out-of-scope:

• Projects focusing purely on machine learning or perception that are unrelated to sensorimotor learning.

If you have a strong reason for pursuing such a project, an exception may be made with instructor’s explicit permission.

Team Size