{"ID":6138825,"CreatedAt":"2026-07-09T01:07:32.349475501Z","UpdatedAt":"2026-07-10T17:41:27.792927618Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2607.06655","arxiv_id":"2607.06655","title":"Pelican-VLA 0.5: Attending Before Acting Benefits Generalization","abstract":"In this report, we present Pelican-VLA 0.5, a unified VLA model that integrates vision-language understanding, future-frame generation, and action prediction within a single architecture. Pelican-VLA 0.5 achieves attention-level generalization: without object annotations, segmentation masks, attention supervision, or task-specific fine-tuning, its action pathway already focuses on the instruction-relevant object and contact region. This behavior persists across unseen scenes and unseen robot embodiments, and is substantially stronger than in other open-source VLA baselines. We verify that this ability originates from the learnable Reasoning Slots inserted between perception and action: by routing task-relevant visual information through a compact bottleneck, the slot interface induces manipulation-centric attention during pre-training and remains effective across different policy structures, including a MoT-style architecture.","short_abstract":"In this report, we present Pelican-VLA 0.5, a unified VLA model that integrates vision-language understanding, future-frame generation, and action prediction within a single architecture. Pelican-VLA 0.5 achieves attention-level generalization: without object annotations, segmentation masks, attention supervision, or t...","url_abs":"https://arxiv.org/abs/2607.06655","url_pdf":"https://arxiv.org/pdf/2607.06655v1","authors":"[\"Zeyuan Ding\",\"Wenhai Liu\",\"Yang Xu\",\"Jiayu Hu\",\"Yinda Chen\",\"Yi Zhang\",\"Yong Dai\",\"Jian Tang\",\"Xiaozhu Ju\"]","published":"2026-07-07T17:50:22Z","proceeding":"cs.RO","tasks":"[\"cs.RO\",\"cs.LG\"]","methods":"[]","has_code":false}