A universal planning system for hybrid domains

Abstract

Many real world problems involve hybrid systems, subject to (continuous) physical effects and controlled by (discrete) digital equipments. Indeed, many efforts are being made to extend the current planning systems and modelling languages to support such kind of domains. However, hybrid systems often present also a nonlinear behaviour and planning with continuous nonlinear change that is still a challenging issue.In this paper we present the UPMurphi tool, a universal planner based on the discretise and validate approach that is capable of reasoning with mixed discrete/continuous domains, fully respecting the semantics of PDDL+. Given an initial discretisation, the hybrid system is discretised and given as input to UPMurphi, which performs universal planning on such an approximated model and checks the correctness of the results. If the validation fails, the approach is repeated by appropriately refining the discretisation.To show the effectiveness of our approach, the paper presents two real hybrid domains where universal planning has been successfully performed using the UPMurphi tool.

CASP solutions for planning in hybrid domains

“…UPMurphi (Della Penna et al 2009;Della Penna et al 2012) is the only tool able to handle the full set of PDDL+ features, although it is very limited in scalability as it performs blind search.…”

Automated planning for Urban traffic control: Strategic vehicle routing to respect air quality limitations

“…The fragment of the corresponding plan (POLLUTING PLAN) focuses on vehicles v1, v2, v3 and v6 and shows the time points at which they traverse the junctions at the borders of the urban area 8 . In the next phase, shown in Figure 2 (b), from POLLUTING PLAN a new problem is generated, with the objective to re-route the vehicles traversing the urban area in case their pollution emissions exceed the safety limit.…”

“…For example, COLIN [6] or a SAT-based approach [19] does only support linear processes, which is hardly the case of the pollution dispersion. UPMurphi [8,9] or dReal [2] are possible options, however, the former performs blind search and thus its scalability is very questionable, while the latter does not support exogenous events. Hence, the absence of effective PDDL+ solvers is the major obstacle in using PDDL+ for modelling our scenario.…”

Automatic Synthesis of Data Cleansing Activities

“…In the AI Planning field a Universal Plan (Schoppers, 1987) is a set of policies, computed off-line, able to bring the system to the goal from any feasible state (the reader can see (Cimatti et al, 1998;Della Penna et al, 2012) for details). Similarly, we are interested in the synthesis of an object, we call Universal Cleanser (UC), which summarises for each pair (state, action) leading to an inconsistent state, the set A ′ of all the feasible cleansing action sequences.…”