MULTI-AGENT TRAJECTORY PREDICTION FOR AUTONOMOUS VEHICLES

Abstract

Autonomous vehicles require motion forecasting of their surrounding multiagents (pedestriansand vehicles) to make optimal decisions for navigation. The existing methods focus ontechniques to utilize the positions and velocities of these agents and fail to capture semanticinformation from the scene. Moreover, to mitigate the increase in computational complexityassociated with the number of agents in the scene, some works leverage Euclidean distance toprune far-away agents. However, distance-based metric alone is insufficient to select relevantagents and accurately perform their predictions. To resolve these issues, we propose theSemantics-aware Interactive Multiagent Motion Forecasting (SIMMF) method to capturesemantics along with spatial information and optimally select relevant agents for motionprediction. Specifically, we achieve this by implementing a semantic-aware selection of relevantagents from the scene and passing them through an attention mechanism to extractglobal encodings. These encodings along with agents’ local information, are passed throughan encoder to obtain time-dependent latent variables for a motion policy predicting the futuretrajectories. Our results show that the proposed approach outperforms state-of-the-artbaselines and provides more accurate and scene-consistent predictions.