RehabMotion-RL-Models
Current clinical/instrumental RL policies: five reward mixtures, five seeds, KL weight 0.01, fixed round 500.
This is a private current-pipeline research backup in the RehabMotion / RL collection.
The VAE, language backbone, clinical regressors, and terminal reference remain frozen. Checkpoints contain the trained policy and frozen terminal reference; do not substitute a different terminal head. The base generator/VAE is the existing private official package RehabMotion-FT-Generator. The exact checkpoint hashes are recorded in the included official registry. This backup includes all five reward mixtures (70/30, 50/50, 30/70, 100/0, and 0/100), with seeds 0 through 4.
Provenance
Source code: https://github.com/LucasbrandaoVIX/RehabMotion/tree/65a15d14212de9afa3ca0db66f86c84f0985291c
All authored documentation and descriptive labels are in English. Immutable source configurations, arrays, predictions, hashes, and source identifiers retain their original representation. No scientific training or generation was repeated to prepare this backup. Historical experiments outside the current pipeline are not included.