Most "run it live" buttons in the slides open a browser module on the course site and need no setup. Ten buttons belong to demonstrations that run a seven-joint robot arm in the Drake robotics toolbox. These run in Python on your own computer and show the robot in your browser. They are optional: every number they show is also in the slides' tables and figures. In the slide PDFs, those ten buttons open this page.
Drake installs from PyPI on these systems only:
| Your computer | Python |
| Linux on x86_64 (Ubuntu 22.04 or newer) | 3.10 to 3.13 |
| Mac with Apple silicon (M1 or newer), macOS 15 or newer | 3.12 or 3.13 |
| Windows | use WSL 2 with Ubuntu 22.04 or newer, then follow the Linux steps |
An Intel Mac, or a Mac on macOS 14 or older, has no Drake package; use the browser modules instead.
Download this repository (the green "Code" button on GitHub, then "Download ZIP", or git clone). In a terminal, from the repository's demos folder:
python3 --version # must be one of the versions above
python3 -m venv .venv # a private Python environment for the demos
source .venv/bin/activate # run this again in every new terminal
python -m pip install --upgrade pip
python -m pip install -r requirements.txt
python drake/smoke_test.py # prints the arm's position and Jacobian, then exitsThe first Drake run downloads Drake's robot models once, about 240 MB, which took under a minute on a fast connection; later runs work offline. python drake/smoke_test.py --hold also starts the viewer: open the address it prints in a browser and press Enter in the terminal to quit.
Activate the environment, then run the command listed below from the demos folder. Each command prints an address such as http://localhost:7000 or http://127.0.0.1:52341/; open it in a browser. The demo waits for you and never moves by itself. End it with its "Stop" button, or press Ctrl+C in the terminal. After Ctrl+C, Python prints a few lines ending in KeyboardInterrupt; that is how it reports the stop, not an error. Running a script without its flag prints the numbers that appear on the slide, then exits.
Slides: slides/02-D1-learning-overlay-and-frames.pdf, page 8 (Computed: how much the mug moves per joint).
python drake/00-contact-changes-the-map/contact_demo.py --meshcatSeven joint sliders and a "Grasp / release" button in the browser. Before the grasp, moving the joints leaves the mug where it is; after the grasp, the mug moves with the hand. The terminal prints the rank of the mug's Jacobian after each press.
Slides: slides/02-D1-learning-overlay-and-frames.pdf, page 9 (Challenge 2: state estimation).
python drake/00b-when-the-map-is-wrong/map_error_demo.py --meshcatSliders set the error of the estimated mug position in centimetres. "Plan on the estimate, execute on the truth" plans for the orange ghost mug and executes against the real teal mug, so a small error lands a finger on the rim. "Back to the ready pose" resets the arm.
Slides: slides/02-D2-transforms-fk-jacobian.pdf, page 19 (The FR3 uses the same transform chain).
python drake/01c-fr3-grasp-chain/fr3_grasp_chain.py --meshcatThe Franka Research 3 model in its ready pose. Buttons step through the chain from the base to the flange, the tool center and the mug grasp; sliders change the link and mug opacity and nudge each joint.
Slides: slides/02-D2-transforms-fk-jacobian.pdf, page 25 (Live demo: the Panda's 7-joint arm in Drake); slides/02-D2-transforms-fk-jacobian.pdf, page 57 (Backup: the Panda's zero configuration is singular).
python drake/01-fk-jacobian/fk_jacobian_demo.py --meshcatThe Panda with seven joint sliders. The terminal prints the end-effector position, the rank of the Jacobian and its smallest singular value as you move them. Three buttons set a spread pose, a near-singular pose and the singular zero configuration.
Slides: slides/02-D2-transforms-fk-jacobian.pdf, page 42 (Live demo: redundancy, the elbow swings, the hand stays).
python drake/01b-redundancy-self-motion/self_motion_panda.py --meshcatThe gripper stays on the mug while one slider moves the elbow along the curve of configurations that keep the hand still. Buttons jump to elbow up and the two elbow-down poses, and play a sweep.
Slides: slides/02-D2-transforms-fk-jacobian.pdf, page 43 (Same demo, only the fingertip position fixed).
python drake/01b-redundancy-self-motion/tip_only_panda.py --appA web page with the scene and a readable panel. Only the fingertip position is fixed, so the seven joint sliders (in the scene's Open Controls menu) move the arm in four free directions.
Slides: slides/03-D1-optimization-1.pdf, page 31 (A real arm: the Panda in Drake).
python drake/02-gauss-newton-ik/gn_ik_panda.py --replay --appA web page, paused at the start. Step, Back and the slider walk through the Gauss-Newton iterates; the predicted and the actual flange positions are drawn at each step, and four runs can be selected.
Slides: slides/04-D1-constraints-and-komo.pdf, page 18 (Live demo: IK as a nonlinear program on the Panda).
python drake/03-ik-as-nlp/ik_nlp_panda.py --meshcatTarget sliders in the browser. After each move, the arm re-solves inverse kinematics as a nonlinear program, and the terminal prints which constraints are active.
Slides: slides/04-D1-constraints-and-komo.pdf, page 24 (The augmented Lagrangian learns the multiplier).
python drake/04-constrained-ik-al/constrained_ik_panda.py --replay --appA web page, paused at the start. Step through the augmented-Lagrangian loop with one elbow stop: joint updates move the arm, and multiplier updates change the multiplier with the arm fixed.
Slides: slides/04-D1-constraints-and-komo.pdf, page 27 (The target is a constraint; posture is the cost).
python drake/05-ik-nlp-al/ik_nlp_al_panda.py --replay --appA web page, paused at the start. The target is a constraint and the posture is the cost; step through joint updates and multiplier updates for two targets, and watch a joint bound become active in the second.
pip finds no matching distribution for drake. Your Python version or system is not in the table above. Check python --version inside the activated environment.127.0.0.1 in place of localhost. Under WSL, open it in your Windows browser.The course tested these steps on macOS 26 with Python 3.12 and Drake 1.46.0. Linux and WSL follow Drake's own packages, which the course did not test separately.