Discrete GCBF Proximal Policy Optimization for Multi-agent Safe Optimal Control
Authors: Songyuan Zhang, Oswin So, Mitchell Black, Chuchu Fan
ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We empirically validate our claims on a suite of multi-agent tasks spanning three different simulation engines. The results suggest that, compared with existing methods, our DGPPO framework obtains policies that achieve high task performance (matching baselines that ignore the safety constraints), and high safety rates (matching the most conservative baselines), with a constant set of hyperparameters across all environments. |
| Researcher Affiliation | Academia | Department of Aeronautics and Astronautics, MIT MIT Lincoln Laboratory EMAIL EMAIL |
| Pseudocode | No | The paper presents a diagram in Figure 1 labeled "DGPPO algorithm" but does not include structured pseudocode or an algorithm block. |
| Open Source Code | Yes | The code of our algorithm and the baselines are provided in the dgppo.zip file in the supplementary materials and online at https://github.com/MIT-REALM/dgppo. |
| Open Datasets | Yes | Environments. We evaluate DGPPO in a wide range of environments including four Li DAR environments (TARGET, SPREAD, LINE, BICYCLE) where the agents use Li DAR to detect obstacles (Keyumarsi et al., 2023), one Mu Jo Co environment TRANSPORT (Todorov et al., 2012), and two VMAS environments (TRANSPORT2, WHEEL) (Bettini et al., 2022; 2024). |
| Dataset Splits | No | The paper mentions evaluating each run on "32 different initial conditions" but does not specify traditional training, validation, or test dataset splits, as the data is generated within simulation environments rather than being a static dataset. |
| Hardware Specification | Yes | The experiments are run on a 13th Gen Intel(R) Core(TM) i7-13700KF CPU with 64GB RAM and an NVIDIA Ge Force RTX 4090 GPU. |
| Software Dependencies | No | The paper mentions using JAX (Bradbury et al., 2018) for implementing baselines but does not provide specific version numbers for JAX or other key software dependencies. |
| Experiment Setup | Yes | In Table 1, we provide the value of the common hyperparameters for DGPPO and the baselines. Besides these common hyperparameters, the value of the unique hyperparameters of DGPPO are provided in Table 2. |