AGIdex
SUB-CAPABILITY · EMBODIED COGNITION

Failure Recovery

22%
vs. HUMAN BASELINE = 100
UnsolvedConf · Low

Scoring rubric

0–20
Early research
20–40
Narrow benchmark competence
40–60
Strong benchmark · reliability gaps
60–80
Human-competitive in common scenarios
80–100
Comparable to typical skilled adult
100+
Reliably exceeds typical human baseline

Score vs. baseline

Trend · Last 12 months
12mo ago   %
6mo ago   %
Today   %
Δ 12mo   +0

What this measures

Human baseline

Recognize when a step has failed and stop rather than continuing blindly.

Human frontier

Diagnose why a step failed and replan from current state without human intervention.

Current state

What works

Some VLM-based systems detect obvious failures post-hoc in structured tasks.

Key gaps

Real-time failure detection and autonomous replanning mid-task in unstructured environments.

Evidence

0 sources
TechnologyQualitySourceScore vs. baselineScore
No evidence records yet.
Source: arxiv.org/abs/2510.17801 · Last reviewed May 2026