Weakly-Supervised Multi-Granularity Map Learning for Vision-and-Language Navigation
Authors: Peihao Chen, Dongyu Ji, Kunyang Lin, Runhao Zeng, Thomas Li, Mingkui Tan, Chuang Gan
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show our method outperforms the state-of-the-art by 4.0% and 4.6% w.r.t. success rate both in seen and unseen environments, respectively on VLN-CE dataset. |
| Researcher Affiliation | Collaboration | Peihao Chen1,6 Dongyu Ji1 Kunyang Lin1,2 Runhao Zeng5 Thomas H. Li6 Mingkui Tan1,7 Chuang Gan3,4 1South China University of Technology, 2Pazhou Laboratory, 3MIT-IBM Watson AI Lab, 4UMass Amherst, 5Shenzhen University, 6Information Technology R&D Innovation Center of Peking University, 7Key Laboratory of Big Data and Intelligent Robot, Ministry of Education |
| Pseudocode | No | The paper describes the computational steps and processes using natural language and mathematical equations, but it does not include any formally structured pseudocode or algorithm blocks. |
| Open Source Code | No | We will publish our code upon acceptance. |
| Open Datasets | Yes | We conduct our experiments on VLN-CE dataset, which contains 16,844 path-instruction pairs from 90 scenes in Matterport3D. |
| Dataset Splits | Yes | The dataset is split into the train, seen validation, unseen validation, and test set. |
| Hardware Specification | Yes | We distribute training over 2 NVIDIA V100 GPUs for 3 days on average. |
| Software Dependencies | No | The paper mentions 'Pytorch framework' and 'Habitat 3 simulator' but does not specify their version numbers or other software dependencies with specific versions. |
| Experiment Setup | Yes | We use the same set of hyperparameters as used in the VLN-CE [32] and show these values in Appendix. A FORWARD action moves the agent forward by 0.25 meters and a TURN action turns by 15 . The map size m is set to 100. α, β, and γ in Equation (6) are set to 10 such that four reward terms are in the same order of magnitude at initialization. Progress threshold λp is set to 0.8. We update the waypoint every 3 time steps. |