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 [1].
No RL, No Simulation: Learning to Navigate without Navigating
Authors: Meera Hahn, Devendra Singh Chaplot, Shubham Tulsiani, Mustafa Mukadam, James M. Rehg, Abhinav Gupta
NeurIPS 2021 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present NRNS as a strong baseline for any future image-based navigation tasks that use RL or Simulation. [...] 5 Experiments [...] Tables 1, 2 show the performance of our NRNS model and relevant baselines on the test splits of the Gibson and Matterport datasets. [...] NRNS Module Ablations. Tab. 3 reports detailed ablations of NRNS on the Gibson dataset |
| Researcher Affiliation | Collaboration | Meera Hahn Georgia Institute of Technology Devendra Chaplot Facebook AI Research Shubham Tulsiani Facebook AI Research Mustafa Mukadam Facebook AI Research James M. Rehg Georgia Institute of Technology Abhinav Gupta Facebook AI Research |
| Pseudocode | Yes | Algorithm 1: NRNS Image Navigation |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code for the described methodology or a link to a code repository. |
| Open Datasets | Yes | We evaluate on the standard test-split for both the Gibson [41] and Matterport3D (MP3D) [42] datasets. [...] Gibson: http://svl.stanford.edu/gibson2/assets/GDS_agreement.pdf Matterport3D: http://kaldir.vc.in.tum.de/matterport/MP_TOS.pdf |
| Dataset Splits | No | The paper mentions '1k training instances per scan' but does not specify the overall training/validation/test splits (e.g., percentages or counts) for the datasets used. |
| Hardware Specification | No | The paper mentions '8 GPUs and 16 processes per GPU' and 'single GPU' for training time but does not specify the exact models or other hardware details (e.g., CPU, RAM). |
| Software Dependencies | No | The paper does not provide specific version numbers for any software dependencies (e.g., programming languages, libraries, or frameworks). |
| Experiment Setup | Yes | For training we use 8 GPUs and 16 processes per GPU. We ο¬rst train with 1k episodes per house, which is identical to how NRNS is trained, and we train for 10M steps. [...] Training the GD model takes 20 epochs, requiring 8 hours on a single GPU. Training the GT model takes 10 epochs, requiring 4 hours on a single GPU. |