Temporal Planning with Intermediate Conditions and Effects
Authors: Alessandro Valentini, Andrea Micheli, Alessandro Cimatti9975-9982
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We experimentally demonstrate that our approach is far better than standard encodings in PDDL 2.1 and is competitive with other approaches that can (directly or indirectly) represent intermediate action conditions or effects. We experimentally evaluate the proposed technique by implementing it in a planner called TAMER and comparing against state of the art tools. The comparison comprises domains from the literature and domains inspired by industrial projects where time and temporal constraints are key aspects and the use of ICE facilitates the modeling. Our results show that our technique, thanks to the native support for ICE, is significantly faster and is able to solve many more problems than the state of the art tools on the industrially-inspired domains. |
| Researcher Affiliation | Academia | Alessandro Valentini, Andrea Micheli, Alessandro Cimatti Fondazione Bruno Kessler, Trento, Italy {alvalentini, amicheli, cimatti}@fbk.eu |
| Pseudocode | Yes | The paper includes 'Algorithm 1 Initial state computation', 'Algorithm 2 Existing time-point expansion', 'Algorithm 3 Action opening expansion', 'Algorithm 4 Search algorithm', and 'Algorithm 5 Relaxed actions'. |
| Open Source Code | Yes | TAMER and all the benchmarks are available at: https://es-static.fbk.eu/people/amicheli/resources/aaai20. |
| Open Datasets | Yes | In particular, we took all the MAJSP, Temporal IPC and Uncertainty IPC instances (we disregarded the HSP instances that are not directly expressible in a planning problem with ICE). The Temporal IPC class is composed of temporal planning domains (without ICE) of the IPC-14 competition (Vallati et al. 2015) for a total of 98 planning instances. The Uncertainty IPC class consists of the same planning instances where the durations of some actions are assumed to be uncontrollable and the rewriting in (Cimatti et al. 2018) is used to produce equivalent temporal planning problems with ICE. Finally, we added a new domain, called PAINTER: a worker has to apply several coats of paint on a set of items guaranteeing a minimum and a maximum time between two subsequent coats on the same item. We created 300 instances of this domain by scaling the number of coats (from 2 to 11) and items (from 1 to 30) and formulated each instance in both ANML, TPP (the language of TPACK) and PDDL 2.1. TAMER and all the benchmarks are available at: https://es-static.fbk.eu/people/amicheli/resources/aaai20. |
| Dataset Splits | No | The paper does not explicitly state training/validation/test dataset splits with specific percentages or counts. |
| Hardware Specification | Yes | We ran all the experiments on a Xeon E5-2620 2.10GHz with 1800s/15GB time/memory limits. |
| Software Dependencies | No | The paper mentions that TAMER is written in C++ and that the code generator uses the GCC C++ compiler, but it does not provide specific version numbers for GCC or any other software libraries or dependencies. |
| Experiment Setup | No | The paper mentions that TAMER uses the 'standard hadd classical planning heuristic' and a 'best-first tree-search algorithm, using an A -like heuristic schema', along with time/memory limits. However, it does not provide specific hyperparameter values, initialization details, or other system-level training settings beyond these general descriptions for the search algorithm. |