Combining optimization and dynamic movement primitives for planning energy optimal forestry crane motions

Abstract

In timber harvesting in Fennoscandinavia, forestry cranes play an important role in maneuvering logs quickly and safely. However, despite their seemingly simple design, forestry cranes are difficult to operate with joysticks, leading to work that is not often optimal, especially in terms of energy consumption. Therefore, forestry machine manufacturers have begun to focus on automation to facilitate manual operation and aim to have fully autonomous machines in the future. This article presents the development of a motion planner to be used in the context of automating the motions of forestry cranes. The goal is to plan energy optimal motions to improve the machine's energy efficiency when the crane is operating with a feedback motion control system. Our development is based on dynamic movement primitives (DMPs), a machine-learning approach that uses human demonstrations as a way to instruct the algorithm on how to plan motions. To this end, we use data of a professional operator performing regular forwarding tasks in a commercial machine that we equipped with motion sensors. Our contribution is a combination of DMPs with an optimization algorithm that exploits the crane's redundancy to find energy-optimal trajectories after the system has learned from a human. Simulation test results show that DMPs are able to replicate human-like controlled motions, with improvements in the energy of nearly 25%, given the nature of planning smooth trajectories. However, incorporating our energy optimization approach leads to improvements of over 40%.