ME 449 Robotic Manipulation

From Mech
(Difference between revisions)
Jump to: navigation, search

Revision as of 11:03, 15 September 2020

Fall Quarter 2020

In Fall 2020, ME 449 will be taught as a hybrid course. Initially, at least, in-person meeting times will be treated as office hours, with a combination of worked examples by the instructors, questions from the class (online or in person), and time to work on assignments with an instructor available if you get stuck. All formal lectures are pre-recorded for viewing any time, and experiments with robots will be in simulation. You do not have to attend in-person meetings for this course, nor do you have to attend online during the course meeting periods, though that would allow you to interact with me and others, live. Advantages of being in person may be a better spatial awareness of your classmates and the instructor, better access to typical social cues that are hard to get through a computer screen, and, as a result, better engagement with others (while respecting social distancing), but there will be no specific material or experiences available only to those who attend in person.

Important Information

  • Zoom meeting info will be emailed and made available in Canvas shortly. All class sessions will be recorded and available in Canvas afterward.
  • The first two class meetings (Sept 16 and 18) will be online only. We will not meet in LR2 until Monday Sept 21. After that, we will evaluate whether an in-person component adds significant value.
  • For ME PhD students: currently ME 512 has a conflict with ME 449. For now, make sure you sign up for ME 449, and don't worry about signing up for ME 512. Since there are many students with this conflict, we are looking into a way to resolve it for everyone without you having to get virtual signatures on a form.
  • We will use this wiki, the Canvas page, and Coursera extensively in this course.

Getting Started Do the following things as soon as possible:

Class Info

Contents

Course Summary

Representations of the configuration and spatial motion of rigid bodies and robots based on modern screw theory. Forward, inverse, and differential kinematics. Robot dynamics, trajectory planning, and motion control. Wheeled mobile robots and mobile manipulation.

Prerequisites

Linear algebra, first-order linear ODEs, freshman-level physics/mechanics, a bit of programming background.

Grading

TBD.

Your lowest assignment grade will be dropped. This policy is meant to handle all eventualities and emergencies: travel, job interview, overloaded that week, computer crashed, could not submit on time (even by 30 seconds!), problem with submitting to Canvas, dog ate it, slept in, forgot, busy watching the big game, big party the night before, etc. Please don't ask for an extension or an exception; you've already been granted one! (But only one.)

Course Text and Software

This course uses the textbook Modern Robotics: Mechanics, Planning, and Control, Kevin M. Lynch and Frank C. Park, Cambridge University Press 2017. If you find an error or typo in the book, please report it here.

Get the book, install and test the Modern Robotics code library, and install and test the CoppeliaSim robot simulator. You will program in Python, Mathematica, or MATLAB in this course.

Here is a linear algebra refresher appendix to accompany the book.

Approximate Syllabus and Reading

  • Chapter 2, Configuration Space (weeks 1-2)
  • Chapter 3, Rigid-Body Motions (weeks 2-3)
  • Chapter 4, Forward Kinematics (week 4); section 4.2 is optional
  • Chapter 5, Velocity Kinematics and Statics (week 5)
  • Chapter 6, Inverse Kinematics (week 6); focus on section 6.2
  • Chapter 8, Dynamics of Open Chains (weeks 6-7); skip sections 8.4, 8.8, and 8.9
  • Chapter 9, Trajectory Generation (week 8); focus on sections 9.1 and 9.4
  • Chapter 11, Robot Control (week 9); focus on sections 11.1 through 11.4
  • Chapter 13, Wheeled Mobile Robots (week 10); skip section 13.3

Video Lectures and the Flipped Classroom

This course will take advantage of video lectures. You can see the video lectures at the video browser http://modernrobotics.northwestern.edu or using direct links to the videos on YouTube. We will be roughly following the associated courses on Coursera.

The general flow of the class will be the following:

  • Before class, watch the videos, take the lecture comprehension quizzes associated with each video, and do the associated reading. In general, I recommend that you first watch the videos to get a quick understanding of the material of the chapter, then follow up by reading the appropriate sections of the book. The videos are short and dense, so don't expect to get by only watching the videos. You will need to read the book, then do the exercises, to gain mastery of the material.
  • In class, I will usually briefly review lecture comprehension quizzes, work an example problem, take questions, and be available as you work on homework.
  • On days when a homework is turned in, I will leave time for any questions about it.

Practice Exercises

Sample exercises and their solutions, useful for practicing your understanding of the material.

Practice Quizzes

Student-Created Exercises

All students will be responsible for creating a practice exercise, consisting of the exercise and the solution. A good exercise should test an important concept in the context of a real robotics application (e.g., motion planning for a quadrotor, robot localization, computer vision, grasping, etc.), require the learner to understand and apply equations in the book or use the book's software, and require a bit of thought (i.e., not just "plug and chug" questions). For many exercises, a good figure or two is helpful. You could use a figure of a real robot and add your own annotations to it (e.g., frames or objects in its environment), or you could hand-draw something, or you could use CoppeliaSim or other software to help create the figure. You should not confine your question to an application discussed in the textbook. Make your exercise interesting and motivating! Exercises that require synthesizing two or more concepts or equations are more interesting and useful. Think about what kind of exercise would have helped you to really understand the material. Your questions should be very clearly worded, so anyone can understand it without you having to be there to interpret it for them.

You should look at the practice exercise document and end-of-chapter exercises for inspiration, but obviously your exercises should not be copies.

You will create your exercise using LaTeX (pronounced "lay teck" or "lah teck"), the standard for scientific document preparation. Overleaf is a free online implementation of LaTeX. To get started on your exercise,

  1. Download this .zip file and uncompress it. There are five files: main.tex, prelims.tex, twist-wrench.pdf, table-lamp.PNG, and LampSolution.PNG.
  2. Create an account on Overleaf.
  3. Create a new (blank) project on Overleaf called "exercise."
  4. Upload the five files to this project. (You may get a warning that your default main.tex file is being overwritten; don't worry about it.)
  5. Click on main.tex to see your main LaTeX document.
  6. Press the "Recompile" button to see the pdf document that is compiled from the five files. You can download the pdf file, or all the "source" files, by clicking on "Menu" and choosing which to download. This is the .pdf file you should have created.

main.tex is the main file of the project, and the only one that you will edit, so you should understand what is going on in that file. prelims.tex tells LaTeX what packages to use and defines some macros, e.g., \twist creates \mathcal{V} and \wrench creates \mathcal{F}. The other three files are image files that get included in the document. You will create different image files depending on your exercise. For example, you can make a nice hand drawing and then scan it.

To learn more about typesetting in LaTeX, google is your friend! Try googling "latex math" or "latex math symbols," for example.

You will turn in the source for your exercise as a zip file, as well as the final pdf file.


Assignments

Assignments are graded based on correctness, how well you organize your homework (it should be easy to understand your thinking and easy to find your responses), and how well you follow the submission instructions below. You will lose points if you don't follow these instructions.

You will not receive credit if you just give an answer. Your solution must demonstrate how you got the answer. It must be easy to follow.

If you ever think a problem is stated incorrectly, not enough information is given, or it is impossible to solve, don't panic! Simply make a reasonable assumption that will allow you to solve the problem (but clearly state what this assumption is), or indicate why it is not possible to solve the problem.

Instructions for uploading assignments to Canvas:

  1. Upload on time! Late submissions are not accepted under any circumstances. See the policy on dropping the lowest assignment grade under "Grading" above.
  2. For every assignment, you should upload exactly one pdf file, named FamilyName_GivenName.pdf. This pdf file should have answers to all the questions, including screen shots, text logs of code running, etc. Always include output of your code running on the exercises, so the grader can see what you got when you ran your code. You may scan handwritten solutions (provided they are neat!), but in any case, all answers should be in a single pdf file. DO NOT UPLOAD SCANS AS JPGS! THEY MUST ALL BE COMPILED INTO A SINGLE PDF FILE.
  3. If required by the assignment (e.g., if you wrote code), in addition to the pdf file above, you should provide a zip file including all source code in their original forms, such as .m, .py, or .nb. This zip file should be named FamilyName_GivenName.zip. Always create a script (for example, titled exercise6-9) that the grader can easily invoke to run your code for a particular exercise. Don't expect the grader to search through your code to find sample code to cut-and-paste. Make it as easy as possible for the grader (you can include a "README.txt" file in your zip file, for example, to tell the grader how everything works).


Final Project: Mobile Manipulation

The final project, described on this page, is due on Canvas on Wednesday during finals week. Reminders:

  1. Read and follow closely the instructions on what to submit! If you are missing requested files, or if you use a different directory structure, you will lose points. Make sure your top-level README file is clear on what you've done and what you've submitted.
  2. If your code does not work well, please describe the remaining issues in your README file. Don't gloss over them or only provide examples where the code works well if the code does not work well for other example problems. Otherwise, if the graders find problems with your software, you will not receive credit for having identified them yourself.
  3. You can get up to 10 pts of extra credit for correctly implementing joint-limit avoidance (so the robot links and chassis do not self-intersect) and singularity avoidance (e.g., using joint limits that keep the arm in a portion of its workspace where it does not encounter any singularities). If you implement these, it is best to submit examples of your code solving the same problem two ways---not using joint-limit avoidance and using it---so the usefulness of the joint-limit avoidance is apparent.
  4. Make sure to keep your problem inputs separate from the code. The exact same code should solve all your problem instances; you shouldn't have different copies of your code for different problem inputs. You could have an input file for each of your examples (e.g., bestScript, overshootScript, newTaskScript) which defines the inputs (e.g., block configurations, controller gains, initial robot configuration) and invokes your code. Then a grader just needs to invoke those scripts to verify your results. (If you implemented joint-limit avoidance, this could just be one of your inputs, e.g., a variable called "avoidJointLimits" which is 0 if you don't care about avoiding joint limits and 1 if you do.)
  5. Make sure your videos are good quality. They shouldn't be too fast (at least 5 seconds long) or low resolution. The motion should be smooth.
  6. If your code is written in python, indicate which version of python should be used.
  7. If you submit your code as part of the MR library, make it easy for the grader to find your code (e.g., collect it all in one place and indicate in the code or your README where to find it).


Personal tools