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; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Fail-Safe Adversarial Generative Imitation Learning
Authors: Philipp Geiger, Christoph-Nikolas Straehle
TMLR 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In an experiment on real-world driver interaction data, we empirically demonstrate tractability, safety and imitation performance of our approach. |
| Researcher Affiliation | Industry | Philipp Geiger EMAIL Bosch Center for Artificial Intelligence Renningen, Germany Christoph-Nikolas Straehle EMAIL Bosch Center for Artificial Intelligence Renningen, Germany |
| Pseudocode | No | The paper only provides a high-level outline of the method in Figure 1, describing the steps in paragraph text and a block diagram, but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Implementation code is available at: https://github.com/boschresearch/fagil. |
| Open Datasets | Yes | We use the open high D data set (Krajewski et al., 2018), which consists of 2-D car trajectories (each 20s) recorded by drones flying over highway sections (we select a straight section with 1500 trajectories). |
| Dataset Splits | Yes | The data set (scene) consists of 1500 tracks (trajectories), of which we use 300 as test set and 100 as validation, and the rest as training set. |
| Hardware Specification | No | The paper mentions 'One full safe set computation with some limited vectorized parts takes around 1.5s on a standard CPU,' but does not provide specific model numbers or detailed hardware specifications for running experiments or training models. |
| Software Dependencies | No | The paper describes the methods and architectures used (e.g., Gaussian policy, conditional normalizing flow, Res Net18, SAC, Wasserstein GANs) but does not specify any software libraries or frameworks with their version numbers required for reproducibility. |
| Experiment Setup | No | The paper provides details on the policy architectures (e.g., number of hidden layers and units, use of Res Net18, leaky ReLUs) and the training algorithm (GAIL with SAC), but does not explicitly list specific hyperparameters such as learning rate, batch size, or number of epochs for the experimental setup. |