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]
GPUDrive: Data-driven, multi-agent driving simulation at 1 million FPS
Authors: Saman Kazemkhani, Aarav Pandya, Daphne Cornelisse, Brennan Shacklett, Eugene Vinitsky
ICLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Using GPUDrive, we train reinforcement learning agents on the Waymo Open Motion Dataset, achieving efficient goal-reaching in minutes and scaling to thousands of scenarios in hours. We open-source the code and pre-trained agents at www.github.com/Emerge-Lab/gpudrive |
| Researcher Affiliation | Academia | 1NYU Tandon School of Engineering 2Stanford University *These authors contributed equally to this work Corresponding authors: EMAIL, EMAIL, EMAIL. |
| Pseudocode | No | The paper describes agent dynamics using mathematical equations in Appendix B but does not contain a clearly labeled pseudocode or algorithm block. |
| Open Source Code | Yes | We open-source the code and pre-trained agents at www.github.com/Emerge-Lab/gpudrive |
| Open Datasets | Yes | Currently, GPUDrive supports the Waymo Open Motion Dataset (WOMD) (Ettinger et al., 2021) which is available under a non-commercial license. |
| Dataset Splits | Yes | We use reinforcement learning to train a set of agents that reach their goals 95% of the time on a subset of 1000 training scenes. |
| Hardware Specification | Yes | All RL experiments in this paper were run on an NVIDIA RTX 4080 or A100. |
| Software Dependencies | No | The paper mentions several software components like 'gymnasium' (Towers et al., 2024), 'torch', 'jax', 'Pufferlib' (Suarez, 2024), 'nanobind' (Jakob, 2022), and 'Stable Baselines 3' (Hill et al., 2018), but does not specify their version numbers, which are required for a reproducible description of ancillary software. |
| Experiment Setup | Yes | The Table below depicts the hyperparameters used to produce the results in Section 4.2. Table 2: Experiment hyperparameters used for comparing the training runs between Nocturne and GPUDrive in Figure 5. The environment configurations are aligned as closely as possible, using the same observations and field of view. |