Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
NeSyPr: Neurosymbolic Proceduralization For Efficient Embodied Reasoning
Authors: Wonje Choi, Jooyoung Kim, Honguk Woo
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate NESYPR on the embodied benchmarks PDDLGym, Virtual Home, and ALFWorld, demonstrating its efficient reasoning capabilities over large-scale reasoning models and a symbolic planner, while using more compact LMs. |
| Researcher Affiliation | Academia | Wonje Choi, Jooyoung Kim, Honguk Woo Department of Computer Science and Engineering, Sungkyunkwan University EMAIL |
| Pseudocode | Yes | Algorithm of NESYPR is in Appendix. |
| Open Source Code | Yes | We include the source codes in the supplementary materials. |
| Open Datasets | Yes | We evaluate NESYPR on PDDLGym [23], Virtual Home [24], and ALFWorld [25], where inputs include observation, goal, and domain knowledge that are specified symbolically. |
| Dataset Splits | Yes | For training, we use a small set of problem instances paired with plans generated by symbolic planners [27]. In PDDLGym, the train sets include 29 instances for Minecraft, 20 for Rearrangement, and 40 for Glib Rearrangement. The test sets contain 389, 400, and 80 instances respectively, all disjoint from the train data. For Virtual Home and ALFWorld, the train sets consist of 77 and 549 instances, respectively. Each test set is split into seen and unseen sets. The seen set contains 112 and 1,509 instances respectively, and shares the same goal as the train set, but varies in object placement and inter-object relations. The unseen set contains 52 and 1,369 instances respectively, and introduces entirely new tasks, not present in both the train and the seen sets. |
| Hardware Specification | No | The paper states 'Further details of the experimental settings are provided in the Appendix.' in Section 5.1 and the NeurIPS checklist confirms 'We include information on the computer resources in Appendix.'. However, the main body of the provided paper text does not explicitly detail specific hardware components (e.g., GPU/CPU models, memory) used for the experiments. |
| Software Dependencies | No | The paper mentions specific LLM models used, such as 'LLa MA-3.21B [73]' and 'Qwen2.5-0.5B [74]', along with their sizes. However, it does not provide explicit version numbers for ancillary software dependencies like programming languages (e.g., Python) or deep learning frameworks (e.g., PyTorch, TensorFlow) required to replicate the experiments. |
| Experiment Setup | No | The paper states 'Further details of the experimental settings are provided in the Appendix.' in Section 5.1. The main body of the provided text does not contain explicit experimental setup details such as concrete hyperparameter values (e.g., learning rate, batch size, number of epochs) or specific optimizer settings. |