Table 3.
Comparison of different search heuristics in RockSample[7,8] environment, using the Blind policy and PBVI as a lower bound.
| Heuristic | Return | EBR (%) | LBI | Belief Nodes | Nodes Reused (%) | Online Time (ms) |
|---|---|---|---|---|---|---|
| Blind: Return:7.35, |Γ| = 1, Time:4s | ||||||
| Satia and Lave | 7.35 ± 0 | 3.64 ± 0 | 0 ± 0 | 509 ± 0 | 8.92 ± 0 | 900 ± 0 |
| AEMS1 | 10.30 ± 0.08 | 9.50 ± 0.11 | 0.90 ± 0.03 | 579 ± 2 | 5.31 ± 0.03 | 916 ± 1 |
| RTBSS(2) | 10.30 ± 0.15 | 9.65 ± 0.02 | 1.00 ± 0.04 | 439 ± 0 | 0 ± 0 | 886 ± 2 |
| BI-POMDP | 18.43 ± 0.14 | 33.3 ± 0.5 | 4.33 ± 0.06 | 2152 ± 71 | 29.9 ± 0.6 | 953 ± 2 |
| HSVI-BFS | 20.53 ± 0.31 | 51.7 ± 0.7 | 5.25 ± 0.07 | 2582 ± 72 | 36.5 ± 0.5 | 885 ± 5 |
| AEMS2 | 20.75 ± 0.15 | 52.4 ± 0.6 | 5.30 ± 0.06 | 3145 ± 101 | 36.4 ± 0.5 | 859 ± 6 |
| PBVI: Return:5.93, |B| = 64, |Γ| = 54, Time:2418s | ||||||
| AEMS1 | 17.10 ± 0.28 | 26.1 ± 0.4 | 1.39 ± 0.03 | 1461 ± 28 | 12.2 ± 0.1 | 954 ± 2 |
| Satia and Lave | 19.09 ± 0.21 | 16.9 ± 0.1 | 1.17 ± 0.01 | 2311 ± 25 | 13.5 ± 0.1 | 965 ± 1 |
| RTBSS(2) | 19.45 ± 0.30 | 22.4 ± 0.3 | 1.37 ± 0.04 | 426 ± 1 | 0 ± 0 | 540 ± 7 |
| BI-POMDP | 21.36 ± 0.22 | 49.5 ± 0.2 | 2.73 ± 0.02 | 2781 ± 38 | 32.2 ± 0.2 | 892 ± 2 |
| AEMS2 | 21.37 ± 0.22 | 57.7 ± 0.2 | 3.08 ± 0.02 | 2910 ± 46 | 38.2 ± 0.2 | 826 ± 3 |
| HSVI-BFS | 21.46 ± 0.22 | 56.3 ± 0.2 | 3.03 ± 0.02 | 2184 ± 33 | 37.3 ± 0.2 | 826 ± 2 |