Vision Dynamics: Environment Modelling, Path Planning and Control Based on Semantic Segmentation

Autonomous driving has been a hot topic in robotics for several years. Two main approaches have been proposed in the scope of solving the autonomous driving challenge: model-based methods relying on trajectory tracking and control and model-free learning based methods, usually directly mapping input signals to control actions. In this paper, we propose a vision-dynamics approach as an enhancement of the classical model-based perception-planning-control architecture, based on a semantic segmentation learning algorithm. We use the output of the learning algorithm in the planning module of our setup as to generate better local trajectories for our controller to track. We deploy our algorithm in a real-world setup, using a Pioneer 3-DX robot that navigates the innards of a park.