Multi-Agents Trajectory Prediction for Autonomous Vehicles with Multi-Modal Predictions.

Abstract

A comprehensive model to predict the motion of multi-agents on the road.

The model is designed to capture social interactions between agents, without relying on the map information.

Each actor is encoded by a Temporal Convolutional Network (TCN) to capture temporal interactions.

Applies a graph convolution method and combines it with multi-head self-attention.

Apply multi-modal predictions to determine the probability of each individual mode.