Can fingers tendons afford hundreds of repetitive movements?

[pattacini]

Is it feasible to perform training for grasping objects by preparing hundreds of supervised trajectories?

The objects are very light and bit elastic. Movements are kind of slow.
Do you think that tendons are usually OK for such amount of repeated movements?

[pattacini]

Performing high number of repetitive experiments on the grasp is perfectly doable. We did it for our tests on power-grasp, for example.

However, in order to preserve the integrity of the tendons throughout the long sessions and prevent the effect of slackness and friction, it is crucial to ensure that the force exerted by the fingers on the objects remain under control.

To this end, two pathways can be pursued:

1. You can tune the trajectories in open-loop fashion for each object, so that the fingers stop precisely at the desired point and the force will not exceed a certain threshold, reasonably specified.
2. You can rely on a closed-loop mechanism that monitors instant by instant a feedback accounting for the contact force at each finger and then triggers asynchronously an immediate stop as soon as a threshold is overcome.

While (1) can be tedious and intrinsically prone to errors, (2) turns to be quite effective and safe.

We have already designed and developed software to implement (2). Relevant API are exposed in the perceptiveModels library, whose online documentation can be browsed here. Two categories of feedback are considered therein: (a) the tactile feedback retrieved from the fingertips and (b) the elastic feedback provided by the springs in the distal joints of the fingers.

In this respect, you might find useful to look at actionsRenderingEngine (aka ARE), available in the icub-main repository. ARE is an example of a module linking against perceptiveModels, thus allowing the user to easily perform fingers trajectories that account for the elastic feedback in a closed-loop form.

[pattacini]

Other (simpler) examples are also:

• tutorial on perceptiveModels.
• tutorial on actionPrimitives.