Energy-efficient motion planning of an autonomous forklift using deep neural networks and kinetic model

Abstract

Autonomous Forklifts (AFs) play a vital role in smart factories, particularly in the transportation of heavy loads. However, their energy consumption poses a significant challenge as they need to operate for extended periods on a single battery charge. Therefore, energy-efficient motion is necessary to raise their availability. The AF’s movement is dynamically determined by a motion planning algorithm within its navigation system. In light of this, this article introduces a strategy to improve energy efficiency during the motion planning phase. This strategy involves a cooperative approach, utilizing the Deep Neural Networks (DNNs) and AF’s kinetic model to achieve this energy-saving goal. Unlike traditional methods that rely solely on the vehicle’s kinematic model, our approach considers an additional factor, incorporating the influence of the vehicle’s kinetic model for a more comprehensive and accurate energy consumption analysis. First, the kinetic model of an AF is developed by considering the effect of the front-powered wheel. Second, the kinetic model is employed within a time-energy optimization technique, aiming to find the AF’s ideal acceleration. This optimization process generates a dataset that covers a range of AF maneuvers and dynamic parameters. Third, a DNNs model is trained using this dataset to predict the optimal acceleration for the AF. Finally, the trained model is integrated into a motion planning algorithm to determine the optimal and acceptable limits for both linear and angular acceleration. Experiments illustrate that the suggested motion planning method can generate trajectories that are both feasible and optimized for energy consumption. This differs significantly from the typical algorithms which generally results in higher energy use by the AF, occasionally leading to the generation of infeasible trajectories.