Gradient-based data valuation (TracIn) outperforms all human-crafted metadata heuristics for ordering curriculum learning in motion planners.
April 2, 2026
Original Paper
Gradient-Based Data Valuation Improves Curriculum Learning for Game-Theoretic Motion Planning
arXiv · 2604.00388
The Takeaway
Reveals that data difficulty for models is nearly orthogonal to human-defined 'interaction difficulty.' Using gradient-similarity to weight training scenarios achieves significantly lower error with better sample efficiency in complex planning tasks.
From the abstract
We demonstrate that gradient-based data valuation produces curriculum orderings that significantly outperform metadata-based heuristics for training game-theoretic motion planners. Specifically, we apply TracIn gradient-similarity scoring to GameFormer on the nuPlan benchmark and construct a curriculum that weights training scenarios by their estimated contribution to validation loss reduction. Across three random seeds, the TracIn-weighted curriculum achieves a mean planning ADE of $1.704\pm0.0