Kin family of datasets

There are eight datastets in this family . They are all variations on the same model; a realistic simulation of the forward dynamics of an 8 link all-revolute robot arm. The task in all datasets is to predict the distance of the end-effector from a target. The inputs are things like joint postions, twist angles, etc. The family has been specifically generated for the delve environment and so the individual datasets span the corners of a cube whose dimensions represent:

- Number of inputs (8 or 32).
- degree of non-linearity (fairly linear or non-linear)
- amount of noise in the output (moderate or high).

More background on the rationale for these choices can be found in the delve task array document.
Detailed documentation on the kin family is also available.

Dataset profile:

Origin: Simulated

Usage: Development

Number of attributes: 9 or 33

Number of cases: 8192

Number of prototasks: 1

Number of methods run on this dataset: 22

Download kin-family

Dataset naming

All datasets in this family have "kin" as the base of their name (KINematics). A dash (-) is appended to this name, followed by:

  1. An integer value signifying the number of input attributes in each case, for example `32'.
  2. One of the characters `f' or `n' signifying `fairly linear' or `non-linear', respectively.
  3. One of the characters `m' or `h' signifying `medium unpredictability/noise' or `high unpredictability/noise', respectively.

Each dataset has a single Prototask called "dist". Within each prototask are tasks with training set sizes from 64 to 1024, increasing by powers of two.

Contributed by: Zoubin Ghahramani, using the Matlab Robotics Toolbox (Release 3). The Matlab Robotics Toolbox was written by Peter I. Corke (1996) and is available freely from The MathWorks or from the author.

Corke, P. I. (1996). A Robotics Toolbox for MATLAB. IEEE Robotics and Automation Magazine, 3 (1): 24-32.

Last Updated 5 October 1996
Comments and questions to: