--
6.4212 Additional Assignment
Students in 6.4212 have one additional assignment. The assignment is to pick some topic relevant to the course and produce a Python notebook tutorial on the subject. The level of exposition should be suitable for anyone taking the undergraduate or graduate versions of the course.. This can be an opportunity to explore some background material that will inform your project, but it can be stand-alone. You should pick a paper, section of a book or on-line lecture to base it on. It could also be based on your own research, but it should be relevant to the course. Make sure you acknowledge your source(s). Plan to limit your time investment to the level of a problem set; not a term project. Focus on providing insight, not on exhaustive exploration.
There are two possible due dates: 10/16 and 11/13
We will make a signup sheet available; there will be roughly half the slots on each date. When you sign up, we will ask for the topic you intend to present; a staff member will follow up if there are questions.
We will provide a “template” notebook in Deepnote that has the basic Drake infrastructure installed; you can “pip install” additional packages as needed (but don’t go overboard). Then, you can hand in a link to that notebook on Gradescope.
Some potential topics for tutorial
These are some suitable topics, but this is not intended to be exhaustive. Please contact a staff member if you want to talk over ideas.
- Illustrate more sophisticated Diff IK via optimization
- Using DiffIK for bi-manual operations/task hierarchy
- Numerical IK methods
- Non-Integrability of DiffIK and how to address it
- Grasp candidates from local curvature (see Ex 5.4)
- Generating grasps analytically vs learned method
- RRT shortcutting vs/plus trajectory optimization (see Ex 6.3)
- Non-holonomic motion planning - differential drive
- ICP with realistic data (we can provide some)
- Enhanced ICP - soft correspondences
- Graph of Convex Sets motion planning
- Examples of force/compliance control
- Explore PID control and why we need all the terms
- Vision foundation models: SAM, Depth-Anything, etc.
- Diffusion policy
- Basic RL methods