A Double-Layer LSTM Model Based on Driving Style and Adaptive Grid for Intention-Trajectory Prediction

Area of Science:

• Autonomous Systems
• Artificial Intelligence
• Robotics

Background:

• Ensuring safety in autonomous vehicles (AVs) is critical for their evolution.
• Precise trajectory prediction is essential for enhancing AV safety and performance in complex environments.
• Conventional prediction methods often fail to account for predicted vehicle behavior and interactions.

Purpose of the Study:

• To introduce a novel double-layer long short-term memory (LSTM) model for accurate autonomous vehicle trajectory prediction.
• To overcome the limitations of existing methods by incorporating driving-style and adaptive grid generation.
• To improve the prediction of vehicle intentions and trajectories in intricate driving scenarios.

Main Methods:

• A novel double-layer LSTM model was developed, integrating convolutional and max-pooling layers for feature extraction.
• Multi-sensor data from perception modules were fused to extract vehicle trajectories.
• Driving-style category values and an improved adaptive grid generation method were incorporated.
• Historical trajectory data and vehicle/lane information were leveraged for dynamic grid adjustment.

Main Results:

• The proposed model demonstrated significantly enhanced representation of vehicle motion features and interactions.
• Experiments on the NGSIM US-101 and I-80 datasets showed superior performance compared to existing benchmarks.
• The model achieved higher intention accuracy and a lower root mean square error (RMSE) over a 5-second prediction horizon.

Conclusions:

• The developed double-layer LSTM model effectively predicts autonomous vehicle trajectories by capturing temporal and spatial features.
• The incorporation of driving style and adaptive grids enhances prediction accuracy and accounts for vehicle interactions.
• The model's stability and effectiveness were verified through rigorous experimentation, offering a promising advancement in AV safety technology.