A Survey on Trajectory-Prediction Methods for Autonomous Driving

Data-Driven Interaction-Aware Behavior Prediction for Autonomous Vehicles

“…probability of changing lanes, as well as future motion (spatial-temporal evolution). While there are several possible proposals for solving this particular problem, including classic physics-based and filtering approaches (Huang et al, 2022, Lefèvre et al, 2014, Mozaffari et al, 2020, the core of this work is based on data-driven methods. Using such methods for behavior prediction problems has only recently gained wide attention, much attributed to increasingly available naturalistic driving data sets (Caesar et al, 2020, Colyar and Halkias, 2007, Halkias and Colyar, 2006, Krajewski et al, 2018, Zhan et al, 2019, but also as a results of increased computational capabilities.…”

“…While this has been widely included in prior research on the topic, there are still opportunities for architectural improvements as new methods emerge. As it has been empirically shown that interaction-aware models are superior in terms of prediction accuracy (Huang et al, 2022), the importance of developing new methods that have the potential to improve these capacities cannot be overstated. It is also interesting to study whether interaction-aware capabilities can be strengthened by multi-agent prediction methods, i.e., methods that predict the behavior of several agents concurrently.…”

“…As the availability of naturalistic driving datasets continues to increase (Caesar et al, 2020, Krajewski et al, 2018, Zhan et al, 2019, deep learning has emerged as a popular approach for predicting the intentions and future motion of road users in traffic (Huang et al, 2022, Mozaffari et al, 2020. Their ability to capture complex relationships makes them attractive choices for the diverse and dynamic nature of traffic scenarios.…”

“…Some of which could potentially be due to sensor imperfections (Almalioglu et al, 2022). The main consideration in this thesis is instead on the uncertainty associated with the unknown intentions of surrounding road users (Huang et al, 2022, Lefèvre et al, 2014, Mozaffari et al, 2020, Yang et al, 2018. The capability of predicting the evolution of a traffic situation is a key property of human drivers and developing a similar functionality for self-driving vehicles is essential for realizing resilient autonomous navigation.…”

Deep learning-based interaction-aware trajectory prediction for autonomous vehicles

“…The reviewed learning-based methods mainly focus on those based on supervised learning since they are most related to this project compared to other learning methods, such as clustering and reinforcement learning. We refer the readers to [11,24,107]…”