Learning-based Preference Prediction for Constrained Multi-Criteria Path-Planning

Kevin Osanlou, Christophe Guettier, Andrei Bursuc, Tristan Cazenave, Eric Jacopin

Keywords: Machine Learning, Constraint Programming, Path-Planning

Abstract: Learning-based methods are increasingly popular for search algorithms in single-criterion optimization problems. On the other hand, there are fewer approaches when dealing with multiple-criteria optimization, while numerous applications exist. Constrained path-planning for Autonomous Ground Vehicles (AGV) is one such application, where an AGV is typically deployed in disaster relief or search and rescue applications in off-road situations. The agent can be faced with the following dilemma : optimize a source-destination path according to a known criterion and an uncertain criterion under operational constraints. The known criterion is associated to the cost of the path, which represents the distance. The uncertain criterion represents the feasibility of driving through the path without requiring human intervention. It depends on parameters such as the physics of the vehicle, state of the explored terrains or weather conditions. In this work, we run simulations offline in which we train a neural network model to predict the uncertain criterion. We integrate this model inside a path-planner which can solve problems online. Finally, we carry out experiments on realistic AGV scenarios which suggest the proposed framework requires human intervention less frequently, trading for a limited increase in the path distance.

Author Identity Visibility: Reveal author identities to reviewers