The final project will be your opportunity to explore some of the topics introduced in the course more deeply. It should have some relation to robotic manipulation and use machine learning in some way. It is perfectly fine (and expected) for the final project to be simulation-only. The project can be the basis of a future conference paper.
You are welcome to work by yourself or in a team of two. We will expect more work from projects with more team members.
Important note: please inform the instructor if you are planning to use this project for multiple classes. If using the project for two classes, for example, we would also expect more work as compared to a single-class project.
Please refer to additional guidelines (and some ideas) here for the final project.
The/ deadlines for the project proposal, mid-term report, and final reports are all due at 11:59pm Pacific Time (on Gradescope).
By October 06, 2023: submit your project proposal.
By November 09, 2023: submit your midterm report. [For this, feel free to email Daniel.]
On November 29, 2023: present your projects in class.
By December 06, 2023: submit your written final report.
Those components will not be graded, and are mainly to make sure you are making progress on an appropriate project. See below for what to include in these milestones.
You are strongly recommended to use an existing manipulation environment in simulation which follows the "Gym" interface (this is standard in reinforcement learning). This will reduce the setup cost and let you get right to the interesting questions for the project. For examples of such manipulation environments, look at the list of papers we have been discussing in class.
Here are some milestones for the project proposal and midterm reports, though this might not be applicable to your project (e.g., if you are not using simulation):
The initial proposal should be 1 page and describe the scope of the project. For this, you should have identified a simulation environment to use for manipulation.
The midterm report should be 1-2 pages. By now you should have the following: (1) a random agent which takes random actions in the environment, and (2) an off-the-shelf, learning-based method running in your environment. Please include some results and/or statistics.
Please talk to the instructor for further discussion and questions as needed.
The project is graded according to:
1/5: quality of the in-class project presentation (see below for details).
1/5: quality of the final written report (see below for details).
3/5: quality of the results.
In the last class you will be giving a project presentation to the class. These are set as 5-minute pre-recorded videos. See Piazza for details.
Grading will be based on the quality of the presentation: is it a clean and polished talk (as we might expect from academic conferences), does it adequately explain the main takeaways, etc.
A key suggestion: try to avoid (long) bulleted lists and use visuals, where possible, in your video.
This should be like a conference paper structured as follows:
See Piazza for the exact LaTeX template; it has a few questions specific to CS 699.
You have 7 pages for the most important written content of your work.
Then starting on page 8 (or earlier, if needed) list the references. You can use multiple pages as needed for cited works.
Supplementary material goes after the references.
For grading purposes, consulting the supplementary material is at the instructor's discretion. Thus, please do not keep any material you want the instructor to read in the supplementary.
Grading will be based on the quality of the report: is it written and structured well (like a good conference paper), does it adequately explain the main takeaways, etc.