WisdomBench-Embodied in Practice: Measuring Learning Ability in Vision-Language-Action Agents on LIBERO

This v2 release updates the LIBERO application paper with separated raw data, evidence index files, and portfolio provenance links (DOI: 10.5281/zenodo.20027295).

The paper applies the WisdomBench-Embodied protocol to Vision-Language-Action agents on the LIBERO benchmark. It evaluates longitudinal learning behavior across repeated rounds, focusing on whether architectural memory, recovery, and failure-processing mechanisms improve learning-from-experience beyond first-attempt capability.

Evidence boundary: the included LIBERO raw file is an empirical WB-E application panel for longitudinal analysis. It is intended to support the paper's learning-after-failure claims, not to replace standard LIBERO SOTA leaderboard reporting.