Lost in Recovery: Demystifying Recovery Behavior of Computer-Use Agents via Goal-Oriented Action Replay
What it is
This paper introduces ReTrace, an evaluation framework for studying how computer-use agents recover from errors in long-horizon tasks. It formalizes 'Transition-Consistent Recovery' and uses goal-oriented action replay to systematically analyze the gap between ideal recovery and current agent performance.
The problem it solves
Computer-use agents are fragile in long-horizon settings where errors accumulate, and existing remedies like environment resets or expensive action replay are infeasible or limited.
What you could use it for
- Evaluating the robustness of computer-use agents in long-horizon workflows
- Benchmarking model-based recovery capabilities against action replay baselines
- Identifying specific failure modes in autonomous digital assistants
Why it matters
Establishes recovery as a fundamental capability for reliable computer-use agents, providing a standardized method to measure and improve error resilience in autonomous systems.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.