AI BriefingNVIDIA Releases AlpaJudge Learned Judge Models for Autonomous Driving Reasoning
·2026.10.30 15:50
Key point
The family includes reference-based and reference-free variants released on Hugging Face under OpenMDW-1.1 and Apache 2.0 licenses.
Details
- NVIDIA released AlpaJudge, a family of learned judge models for evaluating reasoning in autonomous driving.
- The models are available on Hugging Face (nvidia/alpajudge) with weights under the OpenMDW-1.1 license and source code under Apache 2.0.
- AlpaJudge-RB (reference-based) processes text inputs (question, reference answer, candidate answer) to output a single alignment probability between 0 and 1.
- AlpaJudge-RF (reference-free) processes camera frames, ego vehicle meta-actions, and a candidate rationale to output five rubric probabilities and aggregated scores.
Model Variants
- AlpaJudge-RB operates without images, relying solely on text alignment.
- AlpaJudge-RF incorporates visual data from front-wide and front-tele cameras alongside vehicle action metadata.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.