Researchers have developed Pointing-VLA, a new approach to vision-language-action models that improves their spatial grounding. This means the model can better understand and interact with its environment. The system uses typed hidden-state spatial readouts to predict points, object-functional grounding heatmaps, and visual trajectories without relying on text-based coordinates or opaque action tokens. Pointing-VLA achieves state-of-the-art performance in several tasks and is more efficient than previous methods, reducing controller time by over 20 times. It also improves autonomous robot success rates when integrated with an action policy. --- Why it matters: This matters to engineers because it provides a more efficient and inspectable interface between embodied reasoning and robot execution, which can improve the performance of robots in various tasks. Source: https://arxiv.org/abs/2608.23138