Skip to Content

RLTube: Reinforcement Learning Based Deposition Path Planner for Thin-Walled Bent Tubes with Optionally Varying Diameter Manufactured by Wire-Arc Additive Manufacturing

This study presents RLTube, an algorithm that uses reinforcement learning (RL) to optimise deposition path planning in wire-arc additive manufacturing for the production of thin-walled bent tubes with variable diameters. Unlike traditional methods based on rigid mathematical rules, RLTube exploits greater flexibility, adaptability and efficiency by processing 2D images of tubes without the need for additional feature extraction steps.


Type of Work: Scientific Publication

Main Author: Jan Petrik

Affiliation: ETH Zurich

Co-Authors: Markus Bambach

Date: 3rd February 2024

Journal: Manufacturing Letters

Online: ScienceDirect

RLTube's methodology integrates a reinforcement learning framework using a Proximal Policy Optimisation (PPO) algorithm with a Convolutional Neural Network (CNN) for feature extraction and a Multi-Layer Perceptron (MLP) for decision making, processing 2D images of bent tubes to determine optimal deposition paths. The algorithm's reward function was designed to minimize angular deviation and layer height differences, promoting consistent layering.
The key findings show that RLTube outperforms the Brute Force Approach (BFA) in achieving optimised deposition paths according to pre-defined evaluation criteria, including minimising angular deviations and ensuring uniform layer heights. The results not only demonstrated an improvement in the quality and efficiency of the manufacturing process, but the study also provides a robust evaluation framework for future methodologies to be compared.


  • Image 1
    The RLTube workflow processes an input image using an RLTube agent, which determines actions (layer heights h1 and h2) and estimates the reward. Heights between h1 and h2 are linearly interpolated. The Field of View is 128x128 pixels, highlighted in green for the current and red for the previous.
  • Image 1
    Top row visualizes 2D projection of bent tubes for which a deposition path shall be found. The bottom row demonstrates the deposition path outcomes scheduled by RLTube.