AutoRubric-R1V: Rubric-Based Generative Rewards for Faithful Multimodal Reasoning
This addresses the issue of unreliable reasoning in multimodal AI systems, offering a scalable solution for enhancing faithfulness in complex tasks.
The paper tackles the problem of spurious reasoning in multimodal large language models by proposing AutoRubric-R1V, a framework that integrates reinforcement learning with verifiable rewards and process-level supervision through automatically collected rubric-based generative rewards, achieving state-of-the-art performance on six multimodal reasoning benchmarks and substantially improving reasoning faithfulness.
Multimodal large language models (MLLMs) have rapidly advanced from perception tasks to complex multi-step reasoning, yet reinforcement learning with verifiable rewards (RLVR) often leads to spurious reasoning since only the final-answer correctness is rewarded. To address this limitation, we propose AutoRubric-R1V, a framework that integrates RLVR with process-level supervision through automatically collected rubric-based generative rewards. Our key innovation lies in a scalable self-aggregation method that distills consistent reasoning checkpoints from successful trajectories, enabling problem-specific rubric construction without human annotation or stronger teacher models. By jointly leveraging rubric-based and outcome rewards, AutoRubric-R1V achieves state-of-the-art performance on six multimodal reasoning benchmarks and substantially improves reasoning faithfulness in dedicated evaluations.