Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Reinforcement Learning with Robust Rubric Rewards

About

While Reinforcement Learning with Verifiable Rewards (RLVR) is effective for deterministically checkable tasks, many vision-language tasks are partially verifiable, demanding multi-criteria supervision (e.g., perceptual details, reasoning steps, and constraints). Rubrics provide a natural interface for this fine-grained supervision, but their effectiveness depends on the execution accuracy during online RL. We propose Reinforcement Learning with Robust Rubric Rewards ($\text{RLR}^3$), extending RLVR from task-level verification to criterion-level verification. $\text{RLR}^3$ routes instance-specific rubrics through two execution paths: an LLM-as-an-extractor paired with a deterministic verifier, or an LLM-as-a-Judge for non-verifiable criteria. To ensure faithful scoring, $\text{RLR}^3$ introduce a minimal exposure strategy that masks ground truths from extractors and images from judges. Furthermore, $\text{RLR}^3$ employs hierarchical aggregation to prioritize essential criteria over additional criteria, and mitigates score saturation within rollout groups. Evaluated on Qwen3-VL-30B-A3B across 15 benchmarks, $\text{RLR}^3$ consistently outperforms RLVR, yielding a 4.7-point improvement over the base model and exceeding the official instruct-to-thinking model gap. Controlled audits confirm our deterministic verification and minimal exposure significantly reduce exploitable false positives.

Ya-Qi Yu, Hao Wang, Fangyu Hong, Xiangyang Qu, Gaojie Wu, Qiaoyu Luo, Nuo Xu, Huixin Wang, Wuheng Xu, Yongxin Liao, Zihao Chen, Haonan Li, Ziming Li, Dezhi Peng, Minghui Liao, Jihao Wu, Haoyu Ren, Dandan Tu• 2026

Related benchmarks

TaskDatasetResultRank
Multimodal UnderstandingMMStar
Accuracy76.9
407
Chart Question AnsweringChartQA
Accuracy91.1
371
Visual Question AnsweringRealworldQA
Accuracy77.6
259
Mathematical ReasoningWeMath
Accuracy74.3
225
Visual Question AnsweringSimpleVQA
Accuracy0.546
164
Mathematical ReasoningMathVista mini
Accuracy83.9
135
Mathematical ReasoningDynaMath
Accuracy78.4
127
Document Visual Question AnsweringInfoVQA
Accuracy0.902
85
Mathematical ReasoningMathVerse mini
Accuracy79.6
83
Visual Question Answeringcountbenchqa
Accuracy93.2
37
Showing 10 of 12 rows

Other info

Follow for update