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..
Seeing through Uncertainty: Robust Task-Oriented Optimization in Visual Navigation
Authors: Yiyuan Pan, Yunzhe XU, Zhe Liu, Hesheng Wang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on both unordered and sequential multi-object navigation tasks demonstrate that NEURO establishes So TA performance, particularly in generalization to unseen environments. Our work thus presents a significant advancement for developing robust, generalizable autonomous agents. |
| Researcher Affiliation | Academia | Yiyuan Pan Yunzhe Xu Zhe Liu Hesheng Wang Shanghai Jiao Tong University EMAIL |
| Pseudocode | Yes | Algorithm 1 Neu RO Training Algorithm |
| Open Source Code | Yes | Code: https://github.com/PyyWill/NeuRO |
| Open Datasets | Yes | We evaluate NEURO on our proposed unordered Multi ON and traditional sequential Multi ON tasks, demonstrating superior performance and generalization over state-of-the-art (So TA) network-based approaches. [...] Among benchmark tasks, Multi-Object Navigation (Multi ON) task [23] evaluates an agent s ability to locate multiple objects within 3D environments, requiring sophisticated scheduling strategies across multiple goals. |
| Dataset Splits | Yes | We test on unseen settings with object counts m = 1, 2, 3 across standard train, validation and test splits from U-MON and S-MON tasks. |
| Hardware Specification | Yes | Training was conducted on a single NVIDIA Quadro RTX 8000 GPU. |
| Software Dependencies | No | The paper mentions 'Python library CVXPY' in Appendix F. However, it does not provide specific version numbers for this or any other software dependencies, which is required for a reproducible description. |
| Experiment Setup | Yes | All models were trained for 250,000 updates. [...] For the PICNNs, we apply the following parameters: Table 9: Hyperparameters for the Partially Input-Convex Neural Network (PICNN). Parameter Value Input dimension 32 Convex input dimension 768 Hidden layer dimension 256 Number of layers 3 Output dimension = V Include y in output layer True Feasibility parameter 0.0 |