Toward Efficient and Robust Behavior Models for Multi-Agent Driving Simulation
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914470417137664 |
|---|---|
| author | Konstantinidis, Fabian Sackmann, Moritz Hofmann, Ulrich Stiller, Christoph |
| author_facet | Konstantinidis, Fabian Sackmann, Moritz Hofmann, Ulrich Stiller, Christoph |
| contents | Scalable multi-agent driving simulation requires behavior models that are both realistic and computationally efficient. We address this by optimizing the behavior model that controls individual traffic participants. To improve efficiency, we adopt an instance-centric scene representation, where each traffic participant and map element is modeled in its own local coordinate frame. This design enables efficient, viewpoint-invariant scene encoding and allows static map tokens to be reused across simulation steps. To model interactions, we employ a query-centric symmetric context encoder with relative positional encodings between local frames. We use Adversarial Inverse Reinforcement Learning to learn the behavior model and propose an adaptive reward transformation that automatically balances robustness and realism during training. Experiments demonstrate that our approach scales efficiently with the number of tokens, significantly reducing training and inference times, while outperforming several agent-centric baselines in terms of positional accuracy and robustness. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_05812 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Toward Efficient and Robust Behavior Models for Multi-Agent Driving Simulation Konstantinidis, Fabian Sackmann, Moritz Hofmann, Ulrich Stiller, Christoph Robotics Computer Vision and Pattern Recognition Scalable multi-agent driving simulation requires behavior models that are both realistic and computationally efficient. We address this by optimizing the behavior model that controls individual traffic participants. To improve efficiency, we adopt an instance-centric scene representation, where each traffic participant and map element is modeled in its own local coordinate frame. This design enables efficient, viewpoint-invariant scene encoding and allows static map tokens to be reused across simulation steps. To model interactions, we employ a query-centric symmetric context encoder with relative positional encodings between local frames. We use Adversarial Inverse Reinforcement Learning to learn the behavior model and propose an adaptive reward transformation that automatically balances robustness and realism during training. Experiments demonstrate that our approach scales efficiently with the number of tokens, significantly reducing training and inference times, while outperforming several agent-centric baselines in terms of positional accuracy and robustness. |
| title | Toward Efficient and Robust Behavior Models for Multi-Agent Driving Simulation |
| topic | Robotics Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2512.05812 |