Research map / Image generation / Autoregressive and latent generation
LANGTRAJ: Diffusion Model and Dataset for Language-Conditioned Trajectory Simulation
Wei-Jer Chang, Wei Zhan, Masayoshi Tomizuka, Manmohan Chandraker, Francesco Pittaluga
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 direct language conditioning | LangTraj · InterDrive test | mADE: 2.65 -> 2.29 | Improved |
| add direct language conditioning | LangTraj · ProSim-Instruct-520k test | mADE: 2.89 -> 2.52 | Improved |
| use direct conditioning instead of LLM-based guidance | LangTraj · InterDrive test | mADE: 2.70 -> 2.29 | Improved |
| use direct conditioning instead of LLM-based guidance | LangTraj · InterDrive test | Meta realism: 0.70 -> 0.72 | Improved |
| add direct text conditioning to collision guidance (vs LLM guidance) | LangTraj · Safety-critical InterDrive | Collision rate: 0.33 -> 0.43 | Improved |
| add direct text conditioning to collision guidance | LangTraj · Safety-critical InterDrive | Map realism: 0.70 -> 0.74 | Improved |
| add collision guidance | LangTraj · Safety-critical InterDrive | Collision rate: 0.04 -> 0.41 | Mixed |
| reduce denoising steps from K=100 to K=5 (ablation) | LangTraj · WOSAC validation | Meta realism: 0.68 -> 0.68 | No clear effect |
| add closed-loop training without teacher forcing (ablation) | LangTraj K=5 · WOSAC validation | Kinematic realism: 0.41 -> 0.38 | Hurt |
| add teacher forcing to closed-loop training (ablation) | LangTraj · WOSAC validation | Map realism: 0.72 -> 0.79 | Improved |
| add closed-loop training with teacher forcing (ablation) | LangTraj · WOSAC validation | Meta realism: 0.68 -> 0.70 | Improved |
| raise CFG weight from 0.0 to 2.0 (ablation) | LangTraj · CFG weight sweep | Map realism: 0.81 -> 0.73 | Hurt |
Closest papers by meaning
- Language-Driven Interactive Traffic Trajectory Generation (NeurIPS 2024 · similarity 0.87)
- Diffusion-ES: Gradient-free Planning with Diffusion for Autonomous and Instruction-guided Driving (CVPR 2024 · similarity 0.87)
- SceneDiffuser: Efficient and Controllable Driving Simulation Initialization and Rollout (NeurIPS 2024 · similarity 0.87)
- DiffusionDrive: Truncated Diffusion Model for End-to-End Autonomous Driving (CVPR 2025 · similarity 0.86)
- Causal Composition Diffusion Model for Closed-loop Traffic Generation (CVPR 2025 · similarity 0.86)
- MAPLE: Latent Multi-Agent Play for End-to-End Autonomous Driving (NeurIPS 2026 · similarity 0.86)
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.
