<p>This paper presents agoraphilic-3D net (A3D-Net), a novel deep learning-based framework for processing 3D point cloud data in mapless path planning within unstructured terrain environments. The proposed method predicts attractive free space forces by directly analyzing unordered 3D point cloud data using the PointNet architecture, eliminating the need for structured or pre-processed inputs. These predicted forces are integrated into the agoraphilic path planning strategy to compute navigation parameters in real time. In this technique, an automated data labeling is introduced to facilitate effective model training. This system evaluates terrain traversability based on slope and surface geometry, enabling scalable generation of ground-truth labels with minimal manual effort. The proposed method consistently outperforms existing mapless path planning algorithms in terms of success rate, path efficiency, and navigation time, while demonstrating better generalization across diverse previously unseen environments. These findings confirm that A3D-Net can reliably operate in unfamiliar environments and highlight its suitability for deployment in practical robotics applications. Overall, A3D-Net offers a robust and scalable solution for navigation in complex, unstructured environments, while processing the unstructured 3D data in real time, making it well-suited for a wide range of industrial use cases.</p>