Search papers, labs, and topics across Lattice.
This paper introduces PointRL, a reinforcement learning framework that effectively learns point-level vision-language grounding by leveraging heterogeneous annotation evidence. By converting various forms of annotations into actionable pointing instructions and employing a deterministic checker for scoring, PointRL addresses the challenges of non-unique supervision in visual grounding tasks. The framework significantly enhances the accuracy of the Qwen3.5-4B model on the PointArena benchmark, achieving an improvement from 56.11% to 65.58%, and demonstrates consistent gains across multiple external benchmarks.
PointRL transforms complex visual grounding tasks by using verifiable annotation evidence, leading to a remarkable accuracy boost in spatial understanding.
Vision-language models (VLMs) increasingly rely on point coordinates as a compact and executable interface for visual grounding in GUI interaction, robotic manipulation, and interactive visual systems. However, learning reliable pointing behavior remains difficult because the supervision space is inherently non-unique: many coordinates may be valid within the same target region, while multi-instance instructions require target coverage, count consistency, and duplicate suppression. This work presents PointRL, a verifiable reinforcement learning framework that learns point-level grounding from existing heterogeneous annotation evidence. PointRL converts bounding boxes, masks, and instance labels into pointing instructions, while retaining their target supports, instance membership, and set constraints as hidden verifier evidence, i.e., annotations kept outside the prompt and used by a deterministic checker to score predictions. The proposed reward evaluates parseability, point validity, instance coverage, cardinality consistency, and redundant or missing predictions. On PointArena, PointRL improves the overall accuracy of Qwen3.5-4B from 56.11% to 65.58%. Further evaluations on RoboSpatial, BLINK, and Ref-Adv show same-backbone gains on the evaluated external benchmarks, suggesting that verifiable point-level feedback may benefit spatial grounding in these settings.
