Diversity Metrics¶

Use diversity metrics when multi-modal predictions should cover several plausible futures instead of collapsing to near-identical trajectories.

Reference¶

Standard pairwise trajectory diversity metric.

Quick Example¶

import numpy as np
from robometrics import trajectory_diversity

predictions = np.array([[[0.0, 0.0], [1.0, 0.0]], [[0.0, 0.0], [0.0, 1.0]]])
score = trajectory_diversity(predictions)
print(score)

Metrics¶

behavioral_diversity(behaviors, max_pairs=10000, normalize=False) -> float¶

Formula: mean pairwise Euclidean distance between unique behavior embeddings. Reference: Standard pairwise behavior diversity metric Unit: behavior units Direction: higher is better

Behavioral diversity measures spread across behavior embeddings, trajectories, or action sequences after flattening each sample.

trajectory_diversity(predictions) -> float¶

Formula: mean pairwise ADE between predicted trajectory modes. Reference: Standard pairwise trajectory diversity metric Unit: meters Direction: higher is better

Trajectory diversity measures spread across candidate futures, which is useful alongside accuracy metrics.