Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Visual-Based Forklift Learning System Enabling Zero-Shot Sim2Real Without Real-World Data

Koshi Oishi, Teruki Kato, Hiroya Makino, Seigo Ito

TL;DR

This work addresses the challenge of automating counterbalance forklifts by developing a zero-shot sim2real learning system that relies on a photorealistic CAD-based digital environment and a 1/14-scale physical forklift, avoiding real-world data during training. A vision-based DRL pipeline trains an end-to-end controller for pallet approach, complemented by a supervised loading-decision policy, with training conducted in Isaac Sim under strong domain randomization. Real-world validation demonstrates a 60% pallet-loading success rate and 90% loading-decision accuracy in a safe, 1/14-scale setup, illustrating a practical path toward automation without heuristic hand-tuning. The approach offers a scalable, safety-conscious route to automate counterbalance forklifts and provides a foundation for extending to full-scale systems.

Abstract

Forklifts are used extensively in various industrial settings and are in high demand for automation. In particular, counterbalance forklifts are highly versatile and employed in diverse scenarios. However, efforts to automate these processes are lacking, primarily owing to the absence of a safe and performance-verifiable development environment. This study proposes a learning system that combines a photorealistic digital learning environment with a 1/14-scale robotic forklift environment to address this challenge. Inspired by the training-based learning approach adopted by forklift operators, we employ an end-to-end vision-based deep reinforcement learning approach. The learning is conducted in a digitalized environment created from CAD data, making it safe and eliminating the need for real-world data. In addition, we safely validate the method in a physical setting utilizing a 1/14-scale robotic forklift with a configuration similar to that of a real forklift. We achieved a 60% success rate in pallet loading tasks in real experiments using a robotic forklift. Our approach demonstrates zero-shot sim2real with a simple method that does not require heuristic additions. This learning-based approach is considered a first step towards the automation of counterbalance forklifts.

Visual-Based Forklift Learning System Enabling Zero-Shot Sim2Real Without Real-World Data

TL;DR

This work addresses the challenge of automating counterbalance forklifts by developing a zero-shot sim2real learning system that relies on a photorealistic CAD-based digital environment and a 1/14-scale physical forklift, avoiding real-world data during training. A vision-based DRL pipeline trains an end-to-end controller for pallet approach, complemented by a supervised loading-decision policy, with training conducted in Isaac Sim under strong domain randomization. Real-world validation demonstrates a 60% pallet-loading success rate and 90% loading-decision accuracy in a safe, 1/14-scale setup, illustrating a practical path toward automation without heuristic hand-tuning. The approach offers a scalable, safety-conscious route to automate counterbalance forklifts and provides a foundation for extending to full-scale systems.

Abstract

Forklifts are used extensively in various industrial settings and are in high demand for automation. In particular, counterbalance forklifts are highly versatile and employed in diverse scenarios. However, efforts to automate these processes are lacking, primarily owing to the absence of a safe and performance-verifiable development environment. This study proposes a learning system that combines a photorealistic digital learning environment with a 1/14-scale robotic forklift environment to address this challenge. Inspired by the training-based learning approach adopted by forklift operators, we employ an end-to-end vision-based deep reinforcement learning approach. The learning is conducted in a digitalized environment created from CAD data, making it safe and eliminating the need for real-world data. In addition, we safely validate the method in a physical setting utilizing a 1/14-scale robotic forklift with a configuration similar to that of a real forklift. We achieved a 60% success rate in pallet loading tasks in real experiments using a robotic forklift. Our approach demonstrates zero-shot sim2real with a simple method that does not require heuristic additions. This learning-based approach is considered a first step towards the automation of counterbalance forklifts.

Paper Structure

This paper contains 18 sections, 9 equations, 18 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Vision-based forklift control via DRL
  4. Target task
  5. Policy design for pallet approach
  6. Training method
  7. Loss function
  8. Reward function
  9. Forklift learning system
  10. Digital and real environment
  11. Real forklift
  12. Digital forklift
  13. DRL on Isaac Sim
  14. Decision policy for loading
  15. Real demonstration
  16. ...and 3 more sections

Figures (18)

  • Figure 1: Proposed concept. To implement DRL-based control on a full-scale forklift, we first test sim2real in a smaller environment. Sim2real is applied to the full-scale forklift after confirming the safety of the method.
  • Figure 2: Approach policy design.
  • Figure 3: State used for calculating $R_{+}$.
  • Figure 4: Digital environment
  • Figure 5: Side view (digital)
  • ...and 13 more figures