Forklift automation using a digital twin

A study conducted by Koshi Oishi et al., has been accepted for presentation at the 2025 IEEE International Conference on Robotics & Automation (ICRA2025).
The logistics industry faces significant challenges such as labor shortages and the need for increased efficiency. However, because forklifts operate in diverse environments, full automation poses serious difficulties. In particular, a key challenge has been how to enable machines to learn the operational skills that human operators acquire through training. Moreover, gathering the necessary data for machine learning in real-world settings can be extremely time-consuming and costly.
In this study, we leverage the rapidly advancing technology of digital twins to conduct reinforcement learning solely within a virtual environment, thereby overcoming these hurdles. By integrating digital twins with reinforcement learning, the forklift can autonomously collect and learn from data in the virtual space. We implemented this method on a 1/14-scale forklift in real-world tests, achieving a 60% success rate for pallet-transport tasks. This research serves as a first step toward forklift automation through the fusion of digital twins and machine learning.

Title: Visual-Based Forklift Learning System Enabling Zero-Shot Sim2Real without Real-World Data
Authors: Oishi, K., Kato, T., Makino, H., Ito, S.
Journal Name: IEEE International Conference on Robotics and Automation
Published: 2025年5月20日