A trajectory deformation algorithm for intelligent vehicles

• Ano de defesa: 2023
• Autor(a) principal: Justo, Victor Hugo Sillerico
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Autonomous vehicles; Motion planning; Planejamento de movimentos; Reparação de traje; Trajectory repairing; Veículos autônomos
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-06092023-164648/

Autonomous vehicles require robust planning algorithms to compute the sequence of movements from a starting point to an ending goal while considering the constraints in the environment. It is challenging to ensure safety maneuvers in all possible traffic scenarios and the motion planning module recalculates initially-planned trajectories as many times as necessary to resolve those complex situations. However, the computational cost rises up when the planning process is repeated many times for the same task and current solutions do not allow to link user preferences to the vehicles motion behavior. An alternative is to generate new trajectories based on planned trajectories already available. We propose an algorithm that takes into account Signal Temporal Logic (STL) formulas that represent the constraints imposed by the user in order to modify invalid trajectories and guide the motion planning into respecting safety requirements such as the minimum distance to static obstacles or between vehicles. We use a lattice-based planner to generate candidate paths and include a multi-resolution feature to generate as many lattices as it is necessary depending on the context. Then, the STL robustness value quantifies the level of respect that initial paths have for STL specifications and activates the repairing process that generates new lattices based on the initial selected path. The robustness measure also defines a new resolution to generate lattices and influences the cost function to ensure the selection of the path that has more respect for the STL formulas. The deformed version of the initial lattice is used to generate the trajectory for a specified planning horizon using a simulation approach. The computational cost of the proposed repairing strategy is less than recalculating the complete trajectory from scratch and it is specially convenient when there are not many rule violations near the goal region. We evaluate our approach using the automobile tools of the robot simulator Webots considering different traffic scenarios involving obstacle avoidance. The efficiency of our method is demonstrated by comparing trajectories using STL constraints with trajectories that do not consider STL rules.