The paper presents a Simple and efficient Motion Prediction baseLine (SIMPL) for autonomous vehicles. SIMPL provides real-time, accurate motion predictions for all relevant traffic participants. It does this through a compact and efficient global feature fusion module and continuous trajectory parameterization using Bernstein basis polynomials. This approach mitigates accuracy loss caused by viewpoint shifting and allows evaluations of states and their higher-order derivatives at any desired time point, valuable for downstream planning tasks. The method outperforms others on Argoverse 1 & 2 motion forecasting benchmarks and its lightweight design makes it suitable for real-world onboard deployment.
Publication date: 6 Feb 2024
Project Page: https://github.com/HKUST-Aerial-Robotics/SIMPL
Paper: https://arxiv.org/pdf/2402.02519
