Research map / Architectures and efficiency / Attention and transformers
Efficient Equivariant Transformer for Self-Driving Agent Modeling
Scott Xu, Dian Chen, Kelvin Wong, Chris Zhang, Kion Fallah, Raquel Urtasun
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add distance-awareness features to attention queries and keys (ablation) | DriveGATr 3M · WOSAC 2% validation split | RMM: 0.7408 -> 0.7478 | Improved |
| add invariant adapter after each attention block (ablation) | DriveGATr 3M · WOSAC 2% validation split | minADE: 1.6053 -> 1.5798 | Improved |
| add invariant adapter after each attention block (ablation) | DriveGATr 3M · WOSAC 2% validation split | RMM: 0.7483 -> 0.7478 | No clear effect |
| let each agent attend to 8 nearest map tokens instead of 4 (ablation) | DriveGATr 3M · WOSAC 2% validation split | RMM: 0.7478 -> 0.7528 | Improved |
| let each agent attend to the whole map instead of 8 nearest tokens (ablation) | DriveGATr 3M · WOSAC 2% validation split | RMM: 0.7528 -> 0.7617 | Improved |
| use DriveGATr equivariance instead of explicit RPE | 3M transformer agent model · WOSAC 2024 validation | RMM: 0.7251 -> 0.7620 | Improved |
| use DriveGATr instead of data-augmented non-equivariant Transformer | 3M transformer agent model · WOSAC 2024 validation | minADE: 1.6720 -> 1.4192 | Improved |
| use rotary DRoPE instead of full SE(2) equivariance | 3M transformer agent model · WOSAC 2024 validation | RMM: 0.7206 vs 0.7620 for DriveGATr-3M | Hurt |
| scale DriveGATr from 3M to 30M parameters | DriveGATr · WOSAC 2024 validation | RMM: 0.7620 -> 0.7636 | Improved |
| add explicit pairwise relative positional encodings to Transformer | 3M scene-centric Transformer · WOSAC 2024 validation | RMM: 0.7257 -> 0.7251 | No clear effect |
| add closed-loop fine-tuning (CAT-K) to SMART | SMART-7M · WOSAC 2024 validation | RMM: 0.7678 -> 0.7709 | Improved |
Closest papers by meaning
- Latent Variable Sequential Set Transformers for Joint Multi-Agent Motion Prediction (ICLR 2022 · similarity 0.86)
- TransFuser: Imitation With Transformer-Based Sensor Fusion for Autonomous Driving (TPAMI 2022 · similarity 0.86)
- Scene Transformer: A unified architecture for predicting future trajectories of multiple agents (ICLR 2022 · similarity 0.86)
- DriveTransformer: Unified Transformer for Scalable End-to-End Autonomous Driving (ICLR 2025 · similarity 0.86)
- Real-Time Motion Prediction via Heterogeneous Polyline Transformer with Relative Pose Encoding (NeurIPS 2023 · similarity 0.86)
- BehaviorGPT: Smart Agent Simulation for Autonomous Driving with Next-Patch Prediction (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
