Adaptive interaction-aware trajectory prediction for autonomous vehicles in urban scenarios

• Grant number: 22/04473-0
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Start date: September 15, 2022
• End date: September 14, 2023
• Field of knowledge: Engineering - Electrical Engineering
• Principal Investigator: Denis Fernando Wolf
• Grantee: Iago Pachêco Gomes
• Supervisor: Cristiano Premebida
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Institution abroad: Universidade de Coimbra (UC), Portugal
• Associated to the scholarship: 19/27301-7 - Trajectory and behavior prediction for autonomous vehicles in urban traffic, BP.DR

Abstract

Autonomous vehicles should transform the urban transport scenario, increasing its efficiency, making it more accessible and safe, reducing the environmental impact, among other benefits. To navigate safely, vehicles have algorithms for detecting, classifying and avoiding obstacles. However, due to the dynamics of urban traffic, just detecting the position of an obstacle is not enough to guarantee safety. Thus, multi-target tracking, behavior or intention prediction, and trajectory prediction of these agents are essential and allow decision-making and path planning algorithms to consider likely scenarios, anticipating possible collisions or dangerous situations. The trajectory prediction area is divided into approaches that consider the agents' motion equations, their maneuver intention, and the interaction between the traffic participants. This last case is particularly challenging and is still open in the field of study, due to the complexity of modeling interactions between agents. One of the main challenges is that there are several factors that influence the actions of each driver, for example, psychological factors, driving experience, traffic rules, safety, and actions of surrounding drivers. Thus, the objective of this project is to investigate and develop techniques for interaction-aware trajectory prediction of vehicles that copes with some of its challenges, especially the spatio-temporal interaction and heterogeneity of traffic participant classes (e.g., pedestrian, cyclist, and vehicles). In addition, it is important to take into account that each driver has unique characteristics regarding the execution of traffic participants' actions. Because of this, the prediction framework has an online adaptation module, responsible for adjusting its knowledge to each target agent specificities, thus, continuously improving the performance of the predictor. Finally, this project presents the scope of a research exchange program with the University of Coimbra - Portugal, under the guidance of Prof. Dr. Cristiano Premebida. (AU)