nuScenes Waypoints Model (Trajectory-Only)
Part of the RnB-EnCoRe-SelfDriving collection from the Stanford Autonomous Systems Lab.
- 📄 Paper: https://arxiv.org/abs/2602.08167
- 💻 Code / usage: https://github.com/rnb-encore/RnB-EnCoRe-SelfDriving
This is a Qwen3-VL-4B vision-language model fine-tuned on nuScenes driving data to directly predict a future waypoint trajectory from multi-camera observations, without producing intermediate natural-language reasoning.
For the reasoning + waypoints variant, see stanfordasl/nuscenes-rnbencore-reasoning-waypoints.
Model details
- Base model: Qwen/Qwen3-VL-4B-Instruct (
Qwen3VLForConditionalGeneration) - Architecture: hidden size 2560, 36 layers
- Modality: image/video + text → text
- Task: future trajectory (waypoint) prediction on nuScenes
- Output: predicted waypoints (no chain-of-thought)
Training
- Fine-tuned on a nuScenes VQA trajectory-only dataset
- Epochs: 30 (10,980 optimizer steps)
- Max sequence length: 2048
Usage
For dataset preparation, prompting, inference, and evaluation, follow the instructions in the project repository: https://github.com/rnb-encore/RnB-EnCoRe-SelfDriving
from transformers import AutoModelForImageTextToText, AutoProcessor
model_id = "stanfordasl/nuscenes-waypoints-model"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(model_id, torch_dtype="auto", device_map="auto")
# Build a chat message with the driving camera image(s) + prompt,
# then processor.apply_chat_template(...) and model.generate(...).
# See the GitHub repo for the exact prompt format and post-processing.
Intended use & limitations
This model is a research artifact for autonomous-driving planning experiments. It was trained on nuScenes and is not intended for deployment in real vehicles or safety-critical settings. Outputs may be inaccurate or unsafe; always validate in simulation before any downstream use.
Citation
If you use this model, please cite the RnB-EnCoRe self-driving work: