Skip to content

Latest commit

 

History

History
14 lines (11 loc) · 767 Bytes

File metadata and controls

14 lines (11 loc) · 767 Bytes

Behavioral and generalization evaluation

Use human demonstrations only when their recording metadata includes demonstration_source: human and a pseudonymous participant_id. Create a checksum-bound held-out corpus with create_human_benchmark_manifest.

behavioral_fidelity compares camera velocity, acceleration, jerk, and map occupancy; lower values indicate closer behavior. behavioral_safety_report reports repeated actions, blind fire, and stationary windows. These are evaluation signals, not training rewards or claims of human equivalence.

For every model, plan RGB-only, optional privileged-modality, temporal-context, and leave-one-scenario-out runs separately. Never compare those conditions as though they used identical observations or data.