Target constrained pooling improves trajectory prediction accuracy
Target-Constrained Pooling for Trajectory Extrapolation
Machine Learning
Summary
Predicting the future path of a moving object from limited observations is hard because there are many possible ways it could continue. The authors propose a new method called target-constrained pooling that uses similar past paths to better guess how the path will continue. Their approach combines two ways of looking at the endpoint of movement to make predictions more accurate. When tested in various scenarios, their method significantly reduced prediction errors compared to other methods.
What this means in practice
- •For autonomous vehicle developers: Improve prediction of other vehicles' future paths by combining multiple trajectory signals to reduce errors in tracking and planning.
- •For sports analytics teams: Enhance player movement forecasting by using partial trajectories and related past plays to anticipate future positions more accurately.
Authors
Ning Wang, Claire Cardie, Sainyam Galhotra
Abstract
Trajectory extrapolation from a partial target trajectory is underdetermined, but completed related trajectories provide information about the unresolved continuation. We propose target-constrained pooling (TCP), which separates the finite-horizon endpoint from continuation shape and estimates that endpoint from two complementary source signals. Uniform pooling captures final response levels, while matched-increment pooling captures the remaining change after aligning source and target states. TCP selects representative sources, projects both endpoint estimates onto a target-consistent region, converts them into corrections of the free forecast, and controls their influence through strength calibration and degradation-aware weighting. Across 25,245 targets from five domains and both forecasting directions, TCP reduces mean RMSE by 35.8% to 42.1% for all ten target-only forecasters and improves all 100 forecaster, dataset, and direction comparisons. The two endpoint signals consistently outperform either alone, while representative selection remains within 0.64% to 1.02% of the full donor pool using 12.7x fewer donor inputs and 138.8x fewer matching distance computations.