Estimating arm stiffness by EMG-data

Problem

Defining the Cartesian stiffness matrix of impedance-controlled robots is a quite heuristic task. Additionally, it's necessary to adapt this stiffness matrix to different desired movements and robot tasks. Each human being has been learning to control the limbs' stiffness since birth. As a result, the controlling of stiffness is on a very high level.
For this reason, we want to learn the mechanisms taking place in the human arm and transfer this knowledge to our robot systems.

Solution

To start with experiments on this topic, a system for measuring arm stiffness is needed. To get a maximum freedom of movement and a minimum interference to the human arm system, the measuring will be based on EMG-data. The design and set up of such a system is the goal of this work. Research has already been done on this subject (e.g. N. Hogan, M. Kawato), but there is no system for measuring 3D-stiffness from EMG-signals.

Reference Measurements
To get reference measurements for correlation with EMG-signals, it's necessary to set up a mechanical measurement system. The arm of the experiment's subject will be fixed to DLR's LWR-III (as senn in the figure right). Then, the robot applies positional perturbations to the subject's arm. The bonding of robot and human arm has to be as rigid as possible, but also as secure as possible. The subject always has to be able to free himself from the robot. A force-torque sensor will deliver the necessary data to compute the Cartesian stiffness matrix and EMG-data from chosen muscles will be recorded. If the measurement is too slow, the force response of the human arm will be afflicted with reflexes, which are not part of the stiffness, that the subject had tuned before the robot's interaction. Therefore, the measurements have to be in a very short time interval.

Correlating EMG-signals with the computed stiffness matrix
Stiffness can be considered as a function of EMG-activity. But this functional relation has to be found out. Based on the mechanical measured data, parameter estimation can be done. A suitable method has to be chosen (neural, fuzzy, model based or a combination of these).

Status

The mechanical system for reference measurements is finished and the software for steering exoeriments with the robot is written and has been tested with the robot.

A test-series of mechanical measurements of arm stiffness has successfully been comlpeted. The next step will be to execute more measurements and correlate the meachnical data with the EMG data.