Trajectory Filtering And Quality Based Curriculum Learning

via “trajectory-quality-assessment-and-filtering”

Dataset by nvidia. 3,55,146 downloads.

Unique: Implements multi-modal quality assessment for GR00T-X trajectories (action smoothness, state plausibility, video quality, task completion) with automated filtering recommendations, enabling data-driven dataset curation

vs others: More comprehensive than single-metric filtering because it combines action, state, and video quality signals, and more automated than manual curation because quality assessment is fully algorithmic

via “trajectory filtering and quality-based curriculum learning”

### Other Papers <a name="2023op"></a>

Unique: Applies curriculum learning to trajectory-based policy optimization, enabling agents to learn from mixed-quality data by prioritizing successful examples — this is distinct from uniform trajectory sampling which treats all trajectories equally

vs others: More sample-efficient than uniform sampling because high-quality trajectories contribute more to learning, and more robust than filtering alone because it gradually includes harder cases rather than discarding them

via “adaptive-learning-path-generation”

Unique: Uses learner performance analytics and prerequisite graph algorithms to generate context-aware paths rather than static branching logic; continuously re-optimizes based on ongoing assessment data without requiring manual curriculum redesign

vs others: More granular than Khan Academy's fixed progression model because it adjusts pacing and topic order per-student based on mastery signals, not just completion status