Labference: Benchmarking the Reasoning-Action Connection in Robotic Laboratories
Labference: Benchmarking the Reasoning-Action Connection in Robotic Laboratories
This research presents Labference, a new benchmark for evaluating embodied AI in robotic laboratories. It addresses the gap in existing evaluations by measuring how robots adapt their subsequent actions based on observed physical outcomes rather than just task completion. The study compares different control strategies, revealing that while some models succeed in object transfer, they may still suffer from target-selection errors, whereas others achieve correct selection but face execution failures and high computational costs. Labference aims to make the link between reasoning and action measurable to support autonomous experimentation.
The paper introduces Labference, a benchmark designed to evaluate the connection between physical outcomes and subsequent robotic choices in laboratory settings, distinguishing between reasoning errors and execution failures in current VLA and GPT-driven control policies.
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
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Source：arXiv Robotics — research abstracts · arxiv.org