notesum.ai
Published at November 20Explainable Finite-Memory Policies for Partially Observable Markov Decision Processes
cs.AI
cs.RO
eess.SY
Released Date: November 20, 2024
Authors: Muqsit Azeem1, Debraj Chakraborty2, Sudeep Kanav2, Jan Kretinsky1
Aff.: 1Technical University of Munich, Germany; 2Masaryk University, Brno

| Benchmark | #Nodes in FSC | Stationary Policy size | Transition size | ||||
|---|---|---|---|---|---|---|---|
| #Rows | #DT-nodes | Ratio (FSC/DT-FSC) | #Rows | #DT-nodes | Ratio (FSC/DT-FSC) | ||
| avoid (N=6, R=3) | 9 | 447 | 209 | 2.14 | 121837 | 1381 | 88.21 |
| avoid (N=7, R=4) | 3 | 634 | 233 | 2.72 | 240418 | 7927 | 30.32 |
| evade (N=6, R=2) | 31 | 908 | 613 | 1.48 | 296100 | 14003 | 21.14 |
| evade (N=7, R=2) | 30 | 1085 | 758 | 1.43 | 477329 | 9226 | 51.73 |
| intercept (N=7, R=1) | 5 | 575 | 289 | 1.99 | 176714 | 19551 | 9.04 |
| intercept (N=7, R=2) | 5 | 554 | 255 | 2.17 | 189017 | 11421 | 16.55 |
| obstacle (N=6) | 7 | 22 | 9 | 2.44 | 24 | 9 | 2.67 |
| obstacle (N=8) | 8 | 25 | 10 | 2.50 | 27 | 10 | 2.70 |
| refuel (N=6, E=8) | 5 | 50 | 39 | 1.28 | 555 | 45 | 12.33 |
| refuel (N=7, E=7) | 3 | 24 | 23 | 1.04 | 172 | 15 | 11.47 |
| rocks (N=4) | 52 | 627 | 740 | 0.85 | 22634 | 3578 | 6.33 |