Linear and Integer Programming-Based Heuristics for Cost-Optimal Numeric Planning
Authors: Chiara Piacentini, Margarita Castro, Andre Cire, J. Christopher Beck
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experimental analysis shows that considering an NPHard heuristic often pays off and that A search using our integer programming heuristics establishes a new state of the art in cost-optimal numeric planning. In this section we empirically evaluate the performance of our heuristics. |
| Researcher Affiliation | Academia | Chiara Piacentini, Margarita P. Castro, Andre A. Cire, J. Christopher Beck Department of Mechanical and Industrial Engineering, University of Toronto, Toronto, Canada, ON M5S 3G8 Department of Management, University of Toronto Scarborough, Toronto, Canada, ON M1C 1A4 |
| Pseudocode | No | The paper presents mathematical models and constraints, but it does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any specific link or explicit statement about the release of its source code. |
| Open Datasets | Yes | We consider the set of eight benchmark domains used by Scala et al. (2017): Counters: a set of integer variables X1, ..., Xn can be increased or decreased by actions. The goal is to achieve Xi < Xi+1 for i {1, ..., n} variables (Franc es and Geffner 2015); Gardening: an agent needs to water plants located on a grid. To do so, the agent needs to reach a tap and carry water, subject to a capacity constraint (Franc es and Geffner 2015); Sailing: boats can navigate in an unbounded grid space to rescue people. In order for a boat to rescue a person, it has to reach a region defined by a set of inequalities (Scala, Haslum, and Thi ebaux 2016); Farmland: people can be transported between farms and need to be allocated to given locations (Scala, Haslum, and Thi ebaux 2016); Rover, Satellite, Depots, Zeno Travel: these IPC domains are characterized by resources (energy, fuel, loads) that are produced and consumed, and actions are constrained by these quantities. |
| Dataset Splits | No | The paper refers to benchmark domains but does not provide specific details on training, validation, or test dataset splits. |
| Hardware Specification | Yes | We run all the experiments on a Xeon 3.5GHz processor machine running OS Sierra. |
| Software Dependencies | Yes | The LP and IP models are solved using CPLEX v12.7. |
| Experiment Setup | Yes | We implement the heuristic evaluation in the planner LPRPG (Coles et al. 2008), modified to handle cost-optimal planning by using an A algorithm, where ties are broken preferring states with lower heuristic value. The LP and IP models are solved using CPLEX v12.7. We run all the experiments on a Xeon 3.5GHz processor machine running OS Sierra. Every problem is limited to 4 GB and 30 minutes. |