- 6
- 9 941
Frank Dellaert
Registrace 1. 09. 2008
Statics with Wrenches, Part C: A Half-quadruped
In the last part of what turned out to be a three-part series, I extend the simple wrench recipe to a "half-quadruped" with articulated legs. We recoverer the well known Jacobian transpose relationship between a wrench in the body and the torques in the legs., and I show that the factor graph for this example is quite simple. Finally, I talk a bit about the sticky problem of indeterminacy.
0:00 A cute (half)-quadruped
1:34 The recipe can be 2D or 3D
2:37 Following the recipe
5:13 A Jacobian emerges!
7:00 Creating a "leg factor"
10:59 The full "dog" factor graph
12:59 factor graph code
13:42 Sanity checking FK
16:41 Optimize!
18:34 Discussion
20:00 Indeterminacy
Notebook: colab.research.google.com/drive/17QfCocrG49EVJ15vNg5viWtOXIB48Q5M?usp=sharing
0:00 A cute (half)-quadruped
1:34 The recipe can be 2D or 3D
2:37 Following the recipe
5:13 A Jacobian emerges!
7:00 Creating a "leg factor"
10:59 The full "dog" factor graph
12:59 factor graph code
13:42 Sanity checking FK
16:41 Optimize!
18:34 Discussion
20:00 Indeterminacy
Notebook: colab.research.google.com/drive/17QfCocrG49EVJ15vNg5viWtOXIB48Q5M?usp=sharing
zhlédnutí: 397
Video
Statics with Wrenches, Part B - Hips or Ankles?
zhlédnutí 262Před 3 lety
In this second part, I apply the simple wrench recipe to a four bar mechanism, and I introduce factor graphs to reason about the structure of the equations. With GTSAM, which has all the Lie group math built in, quickly experimenting with different joint configurations and loading factors becomes child's play. 0:00 An animal-like four bar linkage 7:47 Minimizing hip torques with a factor graph ...
Statics with Wrenches, Part A
zhlédnutí 394Před 3 lety
I share a recipe and python notebook to do statics in multi-link systems, as found in robotics, exclusively with wrenches. This is simple, requires little to no creativity, and works every time. 0:00 Intro 1:35 The recipe!!! 5:15 A very simple example 7:24 Adding gravity 9:50 Transforming wrenches 12:18 Some python code 18:28 A two-link arm Notebook: colab.research.google.com/drive/17QfCocrG49E...
Particle Filtering / MCL
zhlédnutí 6KPřed 3 lety
A quick demo of particle filtering, aka Monte Carlo Localization in the context of robotics, with an associated colab notebook which can be found here: colab.research.google.com/drive/1AoGZAFa_8mG1jQAniV1q8bGZsMQnErzl?usp=sharing
Mount Rainier's Eigenvector
zhlédnutí 608Před 3 lety
This video accompanies this post: gtsam.org/2020/08/30/Laplacian.html in which I talk about estimating absolute quantities from relative measurements, using the reconstruction of [Mount Rainier](en.wikipedia.org/wiki/Mount_Rainier) as a motivating example. I'll show how the Hessian of that problem is exactly the "Graph Laplacian" from graph theory, and relate the eigen-decomposition of that gra...
Shonan Rotation Averaging
zhlédnutí 2,3KPřed 4 lety
ECCV Spotlight on Shonan Rotation Averaging (dellaert.github.io/ShonanAveraging/) This is about our ECCV 2020 paper, arxiv.org/abs/2008.02737, "Shonan Rotation Averaging: Global Optimality by Surfing SO(p)^n" Frank Dellaert^1, David M. Rosen^2, Jing Wu1, Robert Mahony^3, and Luca Carlone^2 1 Georgia Institute of Technology, Atlanta, GA 2 Massachusetts Inst. of Technology, Cambridge, MA 3 Austra...
great content Frank, thank you for posting!
thank you for sharing :)
Thanks
Very usefull video, thank you so much! One critique is that youtube put a ton of ads throughout the entire video that really break up the flow, and distract from the subject matter
Many thanks
Thank you!
Really fascinating, thanks for recording and publishing it! I wonder if plotting the next eigenvector for Mount Rainier from above results in a similar split in two halves, but with the yellow line perpendicular to the Fiedler one.
For viewers who want to dig deeper into the formalisms presented in the video: The book "Modern Robotics: Mechanics, Planning, and Control," (Lynch and Park) deals with lie formalism, wrenches and changes of coordinates in wrenches among others. It's easy enough for undergraduates and has an open source library: github.com/NxRLab/ModernRobotics . I did the coursera course that uses the book.
Thank you for sharing.
In the example I use a deterministic motion model, but of course you can also use a stochastic motion model, in which case predictions can be generated by injecting some noise in the forward simulation.
ok, and now try to deploy it on financial market data. Do you think it will be work out?