Do Agentic Policies Need to Look at Every Step? Geometric Abstraction for Reasoning and Action
Do Agentic Policies Need to Look at Every Step? Geometric Abstraction for Reasoning and Action
This paper introduces a framework that shifts routine spatial reasoning from raw visual observations to text and symbols via 3D geometric abstraction. By representing scene geometry through typed, evidence-aware symbolic references grounded in calibrated RGB-D data, the approach allows language models to read compact textual checkpoints with metric residuals after each action. This method enables the reuse of symbolic task structures across decisions while invoking direct visual reasoning only on demand. The authors demonstrate that reasoning over grounded text and symbols improves multi-step manipulation completion and substantially reduces repeated multimodal queries in both offline policy synthesis and continuous online execution, positioning explicit geometric state as an effective substrate for agentic robot policies.
The authors propose a method using geometric abstraction to shift routine spatial reasoning from raw images to text and symbols, aiming to reduce repeated multimodal queries while maintaining multi-step manipulation performance.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Across offline policy synthesis and continuous online execution, reasoning over grounded text and symbols improves multi-step manipulation completion while substantially reducing repeated multimodal queries, positioning explicit geometric state as an effective substrate for agentic robot policies.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org