Beyond Binary Success: A Diagnostic Meta-Evaluation Framework for Fine-Grained Manipulation

2026-05-19

AI Summary

Current embodied intelligence evaluations rely heavily on binary success rates, which fail to uncover fine-grained deficiencies in models’ comprehension, perception, and motor control. This work proposes MetaFine, a diagnostic meta-evaluation framework that unifies heterogeneous benchmarks into multi-difficulty diagnostic scenarios through compositional task graphs, enabling disentanglement of capabilities across three core dimensions and supporting hybrid real-sim validation. For the first time, causal intervention analysis is introduced to identify the local spatial structure preservation capability of visual encoders as a critical bottleneck; enhancing this component significantly improves fine-grained manipulation performance without modifying downstream policies. By shifting evaluation from rank-oriented benchmarking toward actionable diagnostics, MetaFine exposes systematic weaknesses in prevailing vision-language-action (VLA) models and establishes a scalable, highly diagnostic paradigm for future assessment.

Abstract

Fine-grained manipulation marks a regime where global scene context no longer suffices, and success hinges on the tight coupling of local attribute grounding, high-fidelity spatial perception, and constraint-respecting motor execution. However, current embodied AI benchmarks collapse these capacities into binary success rates, systematically inflating reported capabilities by up to 70% and masking the architectural bottlenecks that impede real-world deployment. We introduce MetaFine, a diagnostic meta-evaluation framework that disentangles manipulation competency along three axes: understanding, perception, and controlled behavior. Built on a compositional task graph, MetaFine absorbs heterogeneous external benchmarks and reconstructs them into diagnostic scenarios of varying complexity under a unified protocol. Evaluating state-of-the-art vision-language-action (VLA) models through this lens exposes severe dimension-specific failures invisible to conventional metrics. Through targeted causal intervention, we identify the visual encoder's ability to preserve local spatial structure as a key bottleneck for fine-grained precision: improving it directly unlocks previously inaccessible manipulation capabilities without modifying downstream policies. MetaFine further supports hybrid real-sim validation, using limited paired real-world rollouts to calibrate scalable simulation-based estimates for more stable physical benchmarking. By shifting evaluation from ranking to diagnosis, MetaFine turns benchmarking into an actionable compass for repairing the layered capacities underlying genuine physical dexterity. The MetaFine framework, benchmarks, and supporting resources will be publicly released at our project page: https://metafine.github.io/.

Problem

• fine-grained manipulation
• embodied AI
• binary success rate
• evaluation benchmark
• spatial perception

Innovation

• fine-grained manipulation
• diagnostic evaluation
• vision-language-action models
• spatial perception
• real-sim validation