notesum.ai
Published at December 5Finer Behavioral Foundation Models via Auto-Regressive Features and Advantage Weighting
cs.LG
Released Date: December 5, 2024
Authors: Edoardo Cetin1, Ahmed Touati1, Yann Ollivier1
Aff.: 1University of Paris

| Domain | Task | FB | FB-AW | FB-AWARE (4) | FB-AWARE (8) | LAP-AW |
|---|---|---|---|---|---|---|
| jaco | reach_bottom_left | 49.0±25.5 | 43.9±8.6 | 56.3±8.6 | 63.6±6.4 | 25.9±5.9 |
| jaco | reach_bottom_right | 30.8±7.5 | 71.5±18.2 | 57.6±16.5 | 58.6±21.1 | 34.0±13.9 |
| jaco | reach_random1 | 18.0±8.0 | 42.9±15.7 | 64.4±17.0 | 63.9±10.7 | 20.4±12.6 |
| jaco | reach_random2 | 23.4±6.4 | 55.5±5.6 | 72.8±10.7 | 63.7±8.1 | 14.3±5.1 |
| jaco | reach_random3 | 43.2±27.7 | 39.6±5.9 | 53.1±6.4 | 59.0±12.1 | 14.6±5.6 |
| jaco | reach_random4 | 32.6±23.3 | 57.4±11.5 | 68.4±11.0 | 69.9±10.3 | 24.1±2.8 |
| jaco | reach_top_left | 32.6±12.3 | 41.0±5.4 | 41.9±8.3 | 62.7±14.9 | 10.3±2.2 |
| jaco | reach_top_right | 21.5±11.6 | 25.9±9.2 | 43.6±9.7 | 48.3±12.1 | 21.4±5.2 |
| jaco | Average | 31.4±15.3 | 47.2±10.0 | 57.3±11.0 | 61.2±12.0 | 20.6±6.7 |