Path Planning Transformer Supervised by Improved RRT* with Reduced Random Map Size for Mobile Robots

Lonklang, Aphilak and Botzheim, János (2026) Path Planning Transformer Supervised by Improved RRT* with Reduced Random Map Size for Mobile Robots. INFOCOMMUNICATIONS JOURNAL, 18 (1). pp. 100-108. ISSN 2061-2079

Preview

Text ICJ_2026_1_11.pdf - Published Version Download (1MB) | Preview

Abstract

The Improved Rapidly-exploring Random Tree with Reduced Random Map Size (IRRT*-RRMS) algorithm was previ- ously developed to find collision-free paths for mobile robot path planning. Given the excellent performance of Transformer Neural Networks with sequential data, we propose an encoder-decoder transformer model combining a Vision Transformer (ViT) as the encoder and a time-series forecasting module as the decoder to learn the path planning algorithm. The novelty of this paper lies in developing a model supervised by a dataset generated from the IRRT*-RRMS algorithm and using this trained model for the path planning task. The trained model efficiently predicts intermediate points between the desired starting and goal points. The performance was validated on a real robot, demonstrating that the trained model required less computation time compared to the IRRT*-RRMS algorithm.

Actions (login required)

Edit Item