Search papers, labs, and topics across Lattice.
Pelican-VLA 0.5 is a unified vision-language-action (VLA) model that integrates vision-language understanding, future-frame generation, and action prediction without relying on object annotations or task-specific fine-tuning. The model exhibits attention-level generalization, effectively directing its action pathway towards relevant objects and contact regions even in unseen environments and with different robot embodiments. This capability is attributed to the innovative use of learnable Reasoning Slots, which facilitate manipulation-centric attention by routing visual information through a compact bottleneck, outperforming existing open-source VLA baselines.
Attention-level generalization in Pelican-VLA 0.5 allows it to focus on relevant objects without any task-specific training, outperforming traditional models in unseen scenarios.
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.
