Inverse dynamics benchmark data

This page provides access to data which may be used to test 2-D inverse dynamics software. The data were obtained from a direct dynamics simulation of running, so there is no noise and the intersegmental loads (joint forces and joint moments) are exactly known. Details may be found, until a full paper is written, in an abstract for the 1996 CSB congress in Vancouver ( PDF file or text-only).


Bogert, A.J. van den, and J.J. de Koning, "On optimal filtering for inverse dynamics analysis," Proc. 9th CSB Congress, Burnaby, B.C., pp. 214-215, 1996.

Contributed by: Ton van den Bogert ( Please report problems and errors.

The data files generated by the direct dynamics simulation contain 6001 samples, with sampling interval 0.1 ms. Sample 3001 is heel strike.


These files are available in UNIX format (invdyn_tar.Z, 564 kB) and ZIP format (, 537 kB).

Inertial properties

The 2-D musculoskeletal model had the following inertial properties:

segment       mass [kg]       mom.inertia [kg.m2]    CM to prox joint [% segment length]
THIGH		6.85		0.145361267		0.4323725
SHANK		2.86		0.042996389		0.4334975
FOOT		1.00		0.0200			0.0


A conventional (from the ground up) inverse dynamics analysis can be done using the joint position and ground reaction force data. The results should be the same as the known resultant joint moments and forces.

The idea then is to resample the data files (e.g. use every 50th sample to get 200 samples/s) and add noise, to simulate properties of a typical data collection system. The inverse dynamics analysis is then done on this simulated data set, and differences between the results and the known intersegmental joint forces and moment are studied. The effect of different low-pass filtering strategies will be especially important. When simulating noise, it is important to repeat the procedure several times, without resetting the random number generator, in order to know the reproducibility.

You can use your own 2-D inverse dynamics software, or copy and adapt the following Matlab programs. These programs are free to use, copy and modify, provided that the source is acknowledged. It is the user's responsibility to verify the results of the calculations.