A study case of Dynamic Motion Primitives as a motion planning method to automate the work of forestry cranes

Dynamic motion primitives (DMPs) is a motion planning method based on the concept of teaching a robot how to move based on human demonstration. To this end, DMPs use a machine learning framework that tunes stable non-linear differential equations according to data sets from demonstrated motions. Consequently, the numerical solution of these differential equations represent the desired motions. The purpose of this article is to present the steps to apply the DMPs framework and analyse its application for automating motions of forestry cranes. Our study considers an example of a forwarder crane that has been equipped with sensors to record motion data while performing standard work in the forest with expert operators. The objective of our motion planner is to automatically retract the logs back into the machine once the operator has grabbed them manually using joysticks. The results show that the final motion planner has the ability of reproducing the demonstrated action with above 95% accuracy. In addition, it has also the versatility to plan motions and perform similar action from other positions around the workspace, different than the ones used during the training stage. Thus, this initial study concludes that DMPs gives the means to develop a new generation of dynamic motion planners for forestry cranes that readily allow merging the operator?s experience in the development process.