EPDDL · plank · Aletheia · 16 domains
A planning domain says what agents there are, what can happen, and who observes it. Two questions place every domain of the project: how many agents it reasons about, and whether what they hold is knowledge, which is always true, or belief, which may not be. The answer to the second is not declared. It follows from the kinds of event a domain uses.
Two axes
One equivalence relation. Every event is observed by the only agent there is, so what it holds stays true.
A change the agent does not observe leaves it holding what is no longer the case: the relation loses reflexivity.
One equivalence per agent. Public and semi-private events keep them so, and common and distributed knowledge become definable.
Private and quasi-private events let agents diverge: each keeps what it last had reason to hold, which may now be false.
A domain sits in the belief column as soon as one of its events can go unobserved by an agent it concerns. Public and semi-private events keep every agent's relation an equivalence; private and quasi-private ones do not. The title of each quadrant opens the page that treats it in full, and each name opens that domain's entry.
Solved by the planner
Each domain was grounded by plank and solved by Aletheia on the instance its entry names, with the selection of strategy left to the planner. Depth is that of the returned policy; expansions count the nodes the search expanded to find it.
| tier | domain | frame | agents | atoms | actions | |W0| / |Wd| | goal depth | depth / expansions |
|---|---|---|---|---|---|---|---|---|
| I | lone-inspector | S5 | 1 | 23 | 18 | 2 / 1 | 0 | 5 / 42 |
| I | building-rooms | S5 | 1 | 4 | 4 | 4 / 4 | 1 | 4 / 31 |
| II | remote-door | KD45 | 1 | 9 | 8 | 1 / 1 | 1 | 3 / 14 |
| II | robot-relay | KD45 | 1 | 80 | 133 | 2 / 2 | 1 | 11 / 1 233 |
| III | muddy-children | S5n | 2 | 2 | 2 | 4 / 1 | 1 | 2 / 6 |
| III | Active-Muddy-Child | S5n | 5 | 5 | 5 | 32 / 1 | 1 | none |
| III | robot-warehouse | S5n | 2 | 69 | 132 | 2 / 2 | 1 | replan |
| III | pass-through | S5n | 3 | 16 | 72 | 3 / 3 | 0 | 5 / 445 |
| IV | ugv-coordination | KD45n | 2 | 20 | 17 | 2 / 2 | 2 | 3 / 37 |
| IV | survey-team | KD45n | 3 | 4 | 30 | 2 / 2 | 1 | 3 / 21 |
| IV | survey-sites | KD45n | 3 | 12 | 72 | 3 / 3 | 1 | replan |
| IV | coin-in-the-box | KD45n | 3 | 8 | 21 | 2 / 1 | 1 | 2 / 7 |
| IV | box-task | KD45n | 2 | 10 | 14 | 128 / 8 | 1 | GBFS |
| IV | box-task-2.0 | KD45n | 2 | 14 | 32 | 256 / 4 | 1 | none |
| IV | doxastic-depot | KD45n | 3 | 29 | 315 | 1 / 1 | 1 | 2 / 1 085 |
| IV | hotel-incident | KD45n | 3 | 31 | 840 | 2 / 2 | 1 | 3 / 156 |
replan: solved by replanning over the all-outcomes determinization, which the selection policy prefers for these shapes. GBFS: a linear plan, since the instance needs no sensing. none: the search space was exhausted without a solution; for both such instances the plan stored in the repository is empty as well. Measured with Aletheia as linked by ePlanSys 0.2.0, 45-second budget, on one machine, 25 September 2026.
Next The sixteen domains, one by one Signature, event types, initial model, grounded size, and the goal as written →