Presenter: Francesco Salvetti
Thursday, May 27th, 2021 17:30
Location: Microsoft Teams – click here to join
Francesco Salvetti: Perception and Reasoning: Path Planning Generation with Deep Learning.
Path planning is a fundamental task to achieve robotic autonomous navigation. Given a map of an environment, usually made of an occupancy grid that simply marks navigable locations and obstacles, a path planner algorithm should be able to generate the sequence of coordinates that should be touched by the robotic agent in order to move from a starting place to a given goal. The most known path planning algorithm is the A*, used to find the optimal path between two locations by representing the map as a graph. This algorithm is based on gradual exploration of the environment moving towards the goal, following a heuristic approach to select at each step the most promising node to be explored next. In this talk, a preliminary work on a path planning algorithm based on a Deep Learning network will be presented. The model, inspired by the well-known architecture of the Transformer, usually used for Natural Language Processing tasks, is trained on a synthetic dataset of maze-like maps with the objective of generating the optimal path between two points, with a reduction of the explored nodes. The talk will point out some preliminary results on this topic, as well as identify critical points that must be addressed in the future steps of the research.
Biography: PhD student in Electrical, Electronics and Communications Engineering in collaboration with the two interdepartmental centres Smart Data and PIC4SeR. He got his Master’s Degree in Mechatronics Engineering in 2019 at Politecnico di Torino. He is currently working on Machine Learning applied to Computer Vision and Image Processing in robotics applications.