aiqbench analysis
Pointing-VLA introduces a typed spatial readout architecture that replaces brittle text-based coordinates with direct geometric outputs like heatmaps and normalized points for robot manipulation. This approach significantly accelerates inference speeds and improves real-robot success rates by establishing an efficient, inspectable interface between multimodal reasoning and physical execution. The method demonstrates strong cross-dataset transferability without requiring specific fine-tuning, addressing key bottlenecks in current Vision-Language-Action models.
Pointing-VLA: Typed Spatial Grounding Interfaces for Vision-Language-Action Manipulation
arXiv:2608.23138v1 Announce Type: cross Abstract: Vision-language-action (VLA) models often expose spatial grounding through autoregressive text coordinates or opaque action tokens, creating brittle interfaces between multimodal reasoning and robot execution. We present Pointing-VLA, a typed hidden-state spatial readout built on Embodied-R1. Geometry-specific heads predict normalized points, object-functional grounding (OFG) heatmaps, and visual trajectories without serializing geometry as text. For the evaluated Bridge/WidowX and physical pick-place deployments, an explicit execution contract assigns PICK to source-conditioned OFG and PLACE to Pointing, providing direct stage-aligned spatial targets. Pointing-VLA achieves SOTA performance on Bridge/WidowX, averaging 72.9\% across the evaluated four-task set without Bridge-specific finetuning under collision-enabled CuRobo execution. Pointing and OFG show complementary strengths across native and cross-dataset evaluations. The OFG/contact readout transfers to NORA-1.5, preserving or improving success while reducing recorded controller time by more than 20$\times$; typed heads are also 6.68--6.90$\times$ faster than Embodied-R1 text decoding on a shared external suite. When integrated as spatial guidance for a $\pi_{0.5}$ action policy, Pointing-VLA raises autonomous real-robot success from 52.7\% to 80.7\% across three visual contexts. These results establish typed spatial readouts as an efficient, inspectable interface between embodied reasoning and robot execution.
Abstract:Vision-language-action (VLA) models often expose spatial grounding through autoregressive text coordinates or opaque action tokens, creating brittle interfaces between multimodal reasoning and robot execution. We present Pointing-VLA, a typed hidden-state spatial readout built on Embodied-R1. Geometry-specific heads predict normalized points, object-functional grounding (OFG) heatmaps, and visual trajectories without serializing geometry as text. For the evaluated Bridge/WidowX and physical pick-place deployments, an explicit execution contract assigns PICK to source-conditioned OFG and PLACE to Pointing, providing direct stage-aligned spatial targets. Pointing-VLA achieves SOTA performance on Bridge/WidowX, averaging 72.9\% across the evaluated four-task set without Bridge-specific finetuning under collision-enabled CuRobo execution. Pointing and OFG show complementary strengths across native and cross-dataset evaluations. The OFG/contact readout transfers to NORA-1.5, preserving or improving success while reducing recorded controller time by more than 20$\times$; typed heads are also 6.68--6.90$\times$ faster than Embodied-R1 text decoding on a shared external suite. When integrated as spatial guidance for a $\pi_{0.5}$ action policy, Pointing-VLA raises autonomous real-robot success from 52.7\% to 80.7\% across three visual contexts. These results establish typed spatial readouts as an efficient, inspectable interface between embodied reasoning and robot execution.
Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2608.23138 [cs.RO]
(or arXiv:2608.23138v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2608.23138 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 285 words
Related coverage
VLMs are Native Action Models: Unleashing the Manipulation Capabilities of VLMs with Explicit Action-Space Grounding
arXiv · Oct 12, 2026
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026
ABot-N1: Toward a General Visual Language Navigation Foundation Model
arXiv · Jul 20, 2026
ActionLens: Diagnosing Spatial-Temporal Binding Failures in Vision-Language Models
arXiv · Sep 30, 2026