Interaction-aware trajectory prediction for opponent vehicle in high speed autonomous racing

Abstract

In this paper, we propose an interaction-aware trajectory prediction algorithm considering mutual influence in high-speed autonomous racing. For stable overtaking in 1:N races, it is essential to predict the trajectory of surrounding vehicles taking into account the inter-vehicle effects. To achieve this, we apply the Model Predictive Path Integral technique to trajectory prediction, considering not only information about neighboring vehicles but also prior knowledge about the race track and uncertainty. Model Predictive Path Integral-based prediction is complemented by incorporating a Maneuver Intention Estimation-based trajectory prediction, making it robust to various racing scenarios. To improve trajectory prediction performance, we propose a multimodal perception pipeline that ensures accuracy, reliability, and real-time capability in high-speed environments. The proposed algorithm has been validated in both simulation and real racing tracks, operating at 20ms to ensure real-time performance in high-speed environments.