Active Neural Localization
Authors: Devendra Singh Chaplot, Emilio Parisotto, Ruslan Salakhutdinov
ICLR 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We use a variety of simulation environments for our experiments which include random 2D mazes, random mazes in the Doom game engine and a photo-realistic environment in the Unreal game engine. The results on the 2D environments show the effectiveness of the learned policy in an idealistic setting while results on the 3D environments demonstrate the model s capability of learning the policy and perceptual model jointly from raw-pixel based RGB observations. |
| Researcher Affiliation | Academia | Devendra Singh Chaplot, Emilio Parisotto, Ruslan Salakhutdinov Machine Learning Department School of Computer Science Carnegie Mellon University {chaplot,eparisot,rsalakhu}@cs.cmu.edu |
| Pseudocode | No | The paper describes processes and models but does not include any pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github.com/devendrachaplot/Neural-Localization |
| Open Datasets | No | The paper states, 'We train all the models on randomly generated mazes and test on a fixed set of 1000 mazes (different from the mazes used in training).' It describes the generation process for these mazes but does not provide access information or citation for a publicly available dataset. |
| Dataset Splits | No | The paper mentions training on 'randomly generated mazes' and testing on 'a fixed set of 1000 mazes' or 'a set of 20 mazes', but it does not specify a separate validation set or provide explicit percentages or counts for data splits beyond training and testing. |
| Hardware Specification | Yes | All the 2D experiments (including evaluation runtime benchmarks for baselines) were run on Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz and all the 3D experiments were run on Intel(R) Core(TM) i7-6850K CPU @ 3.60GHz. While all the A3C training threads ran on CPUs, the Unreal engine also utilized a NVidia Ge Force GTX 1080 GPU. |
| Software Dependencies | No | The paper mentions software components like 'Vi ZDoom API', 'Air Sim API', 'A3C', and 'Resnet-18', but it does not specify version numbers for any of these software dependencies. |
| Experiment Setup | Yes | All the models are trained with A3C using Stochastic Gradient Descent with a learning rate of 0.001. We use 8 threads for 2D experiments and 4 threads for 3D experiments. Each thread performed an A3C update after 20 steps. The weight for entropy regularization was 0.01. The discount factor (γ) for reinforcement learning was chosen to be 0.99. The gradients were clipped at 40. |