205 / 2023-10-20 23:10:26

Motion prediction of roadside agents surrounding Autonomous Vehicles on the urban roads

Autonomous Vehicles, Trajectory Prediction, Intent Estimation, Optimal Control, Inverse Optimal Control

6.AI-driven technology > (1)、Intelligent Anomaly Detection, Fault Diagnosis and Prognostics

Abstract: With the rapid development of the application of driverless vehicle technology,

future roads will likely be dominated by commercial and personal autonomous vehicles. By

predicting the future motion of the surrounding agents, Autonomous Vehicles (AVs) will improve

intelligent decision-making and safety. As the safety of Vulnerable Road Users (VRUs) such

as pedestrians, bicycles, etc is often at risk, the precise motion prediction of VRUs in the

local neighbourhood of an AV is essential. The aim of this research is to develop a Motion

Prediction Module (MPM) that could be embedded into AVs and is capable of characterising

the dynamics, estimating intent, and predicting the future motion of the roadside agents in close

proximity from a single-vehicle perspective. Given the recent advances in Inverse Optimal

Control (IOC), a model-based control approach is proposed to estimate the agent’s intentions from

the trajectory observations using the Recovery Matrix IOC method. Subsequently, based on the

recovered intentions future trajectories are predicted using Optimal Control. This research work,

firstly, models the nonlinear pedestrian dynamics as a non-holonomic vehicle and formulates an

objective function that consists of a linearly combined weighted sum of feature functions that

reflect the intentions of pedestrians on the road. Secondly, we develop a preliminary MPM by

implementing a single pedestrian in single-phase motion where the intentions are assumed to

be constant. Thirdly, the preliminary module is tested with synthesised trajectory data and

existing data from the literature.