A Machine Learning Approach for the Three-Point Dubins Problem (3PDP)

We show how problem symmetries plus machine learning can crack the Three-Point Dubins Problem without brute-force enumeration. By mapping all cases to a compact canonical space and learning from 17M+ instances, our models predict the optimal path type with 97.5% top-1 accuracy (100% within top-5) and estimate the intermediate angle within ~2°, delivering fast, reliable solutions for curvature-constrained motion planning in robotics and autonomous systems.

Recommended citation: Saccon, E.; Frego, M. A Machine Learning Approach for the Three-Point Dubins Problem (3PDP). Symmetry 2025, 17, 2133.
Download Paper