MORE | Spring 2025

Multi-agent Interaction-based Trajectory Prediction Using Fused Modalities for Autonomous Vehicles

Safe decision-making is essential in autonomous driving, which requires a holistic representation of the scenario to efficiently predict the trajectory of surrounding agents. Current methodologies widely used in the industry rely on sensors mounted on the ego vehicle, which is prone to blind spots that lead to an incomplete scene representation to make decisions. This research aims to fuse sensors mounted on infrastructure near road scenarios with ego vehicle sensors, which reduces the number of blind spots and increases the look-ahead distance for the vehicle to make safe decisions. In addition, this research aims to fuse the information efficiently to reduce computation and data transmission costs through vector features and data compression.