The field of vision-language models is moving towards improving spatial reasoning capabilities, with a focus on self-supervised learning and reinforcement learning approaches. Researchers are exploring new methods to enhance spatial understanding, including the use of pretext tasks, controllable environments, and viewpoint learning. These innovations have shown promising results in improving model performance on spatial understanding benchmarks and real-world applications. Notably, some studies have demonstrated significant improvements in spatial reasoning abilities, achieving state-of-the-art results on various tasks. Noteworthy papers include: Spatial-SSRL, which introduces a self-supervised RL paradigm that derives verifiable signals directly from ordinary images, resulting in substantial improvements in spatial reasoning. Ariadne, which proposes a controllable framework for probing and extending VLM reasoning boundaries, demonstrating that RL post-training can truly extend the inherent capability boundary of a base VLM. Actial, which presents a two-stage fine-tuning strategy to improve the spatial reasoning capabilities of MLLMs, resulting in significant improvements across multiple tasks. SpatialLock, which proposes a novel framework for precise spatial control in text-to-image synthesis, achieving state-of-the-art results on object positioning tasks.