Search papers, labs, and topics across Lattice.
This paper introduces TRON, a novel online environment for reinforcement learning that generates dynamic training rollouts for visual reasoning tasks, allowing for scalable and verifiable training signals. By utilizing a generator-verifier program, TRON can produce an unbounded stream of tailored instances that adapt to the learner's current curriculum, significantly enhancing the training process. The results show that RL post-training with TRON improves performance across ten external multimodal reasoning benchmarks for various models, demonstrating its effectiveness in fostering better visual reasoning capabilities.
TRON enables an endless supply of tailored training instances, revolutionizing how we approach reinforcement learning for visual reasoning.
Reinforcement learning (RL) for visual reasoning needs scalable, verifiable, and controllable training signals. Existing visual RL post-training trains on static curated datasets, with fixed image-question-answer samples bounded by their collection budget. In this work, we introduce TRON (Targeted, Rule-verifiable Online eNvironments), an online environment substrate: a training rollout is generated on demand by a controllable generator-verifier program that samples a fresh latent visual state, renders an image, asks a question, and exactly verifies the answer. A single run can therefore draw an unbounded stream of fresh instances at the difficulty level required by the current curriculum. The current TRON suite contains 520 environments organized into five ability buckets (spatial, mathematical, diagram, pattern/logic, and counting); the same substrate supports both a single full model trained on all buckets and per-bucket ability-specialist models, with no additional data collection. We also introduce a substrate analysis covering generation reliability, instance and level diversity, cross-environment near-duplicates, and base-model pass rate by difficulty level. RL post-training with METHOD consistently improves performance on ten external multimodal reasoning benchmarks across Qwen3-VL-4B, Qwen2.5-VL-7B, and MiMo-VL-7B-SFT.