Symbol Acquisition for Probabilistic High-Level Planning
Authors: George Konidaris, Leslie Kaelbling, Tomas Lozano-Perez
IJCAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We introduce a probabilistic reformulation of symbolic representations capable of naturally dealing with uncertain representations and probabilistic plans. This is achieved by moving from sets and logical operations to probability distributions and probabilistic operations. We use this framework to design an agent that autonomously learns a completely symbolic representation of a computer game domain, enabling very fast planning using an off-the-shelf probabilistic planner. ... The Treasure Game features an agent in a 2D, 528 528 pixel video-game like world, whose task is to obtain treasure and return to its starting position on a ladder at the top left of the screen (see Figure 2). ... Data was gathered as follows. 100 randomly chosen options were executed sequentially, resulting in one set of data recording whether each option could run at states observed before or after option execution, and another recording the transition data xi = (si, oi, ri, s i) for each executed option. This was repeated 40 times. ... Table 2: Timing results and minimum solution depth (option executions) for example Treasure Game planning problems. |
| Researcher Affiliation | Academia | George Konidaris Leslie Pack Kaelbling Tomas Lozano-Perez Duke University MIT CSAIL Durham NC 27708 Cambridge MA 02139 gdk@cs.duke.edu {lpk, tlp}@csail.mit.edu |
| Pseudocode | No | The paper describes processes in text and shows an example PDDL operator, but does not include structured pseudocode or algorithm blocks labeled as such. |
| Open Source Code | No | The paper does not provide any statement or link indicating that the source code for the methodology is openly available. |
| Open Datasets | No | The paper describes a custom 'Treasure Game domain' and how data was gathered ('100 randomly chosen options were executed sequentially... This was repeated 40 times.'), but it does not provide concrete access information (link, DOI, citation with authors/year) for this dataset. |
| Dataset Splits | Yes | The agent created symbols for each of the partitioned options as follows (all parameters set using 3-fold cross-validation). ... Clustering was performed using the DBSCAN algorithm [Ester et al., 1996] in scikit-learn [Pedregosa et al., 2011], with parameters min samples = 5 and ϵ = 0.4/14 (for partitioning effects) or ϵ = 0.8/14 (for merging start states). |
| Hardware Specification | Yes | Results were obtained on an i Mac with a 3.2Ghz Intel Core i5 processor and 16GB of RAM. |
| Software Dependencies | No | The paper mentions software like scikit-learn and the m GPT planner but does not provide specific version numbers for these dependencies, which are necessary for reproducible descriptions. |
| Experiment Setup | Yes | Clustering was performed using the DBSCAN algorithm [Ester et al., 1996] in scikit-learn [Pedregosa et al., 2011], with parameters min samples = 5 and ϵ = 0.4/14 (for partitioning effects) or ϵ = 0.8/14 (for merging start states). ... The agent created symbols for each of the partitioned options as follows (all parameters set using 3-fold cross-validation). |