Pivot-Centric Trajectory Prediction: Bridging Long Horizons via Dynamical Guidance

The paper introduces Pivot-Centric Trajectory Prediction (PCTP), which decomposes long-horizon motion forecasting into pivot prediction and pivot-based short-term trajectory refinement. Pivot prediction uses global map context and agent-to-agent interactions, while refinement uses local map details to guide shorter trajectory segments. The authors report that PCTP can be integrated with most state-of-the-art predictors and improves accuracy on Argoverse I and II with minimal model-size impact. Combined with QCNet, PCTP reportedly surpassed all published ensemble-free methods on the Argoverse II leaderboard at the time of submission.