Efficient Context-Aware Graph Transformer for Vehicle Motion Prediction

Motion Prediction (MP) of multiple surrounding agents, and accurate trajectory forecasting, is a crucial task for self-driving vehicles and robots. Current techniques tackle this problem using end-to-end pipelines, where the input data is usually a Bird Eye View (BEV) HD map and the past trajectories of the most relevant agents; leveraging this information is a must to obtain optimal performance. In that sense, a reliable Autonomous Driving Stack (ADS) must produce fast predictions. However, despite many approaches use simple ConvNets and LSTMs to obtain the social latent features, State-Of-The-Art (SOTA) models might be too complex for real-time applications when using both sources of information (map and past trajectories) as well as little interpretable, specially considering the physical information. Moreover, the performance of such models highly depends on the number of available inputs for each particular traffic scenario, which are expensive to obtain, particularly, annotated High-Definition (HD) maps. In this work, we propose a transformer-based model that does not rely on HD maps, but on minimal interpretable map information. The proposed model combines the powerful attention mechanisms with GNNs to model agent interactions, it has less parameters than other methods, and it is faster than most previous methods. We achieve near-SOTA results on the Argoverse Motion Forecasting Benchmark. Our code is publicly available at https://github.com/Cram3r95/mapfe4mp.