DART: Semantic Recoverability for Structured Tool Agents
The paper titled 'DART: Semantic Recoverability for Structured Tool Agents' addresses the challenges faced by structured tool agents during execution failures. It introduces a modular runtime called DART that ensures semantic recoverability by localizing failures and certifying restore points. The results demonstrate that DART successfully recovers from commitment-sensitive cases where traditional recovery methods fail.
- ▪DART addresses the dilemma of replaying tasks versus restoring from checkpoints during execution failures.
- ▪The framework certifies semantically recoverable boundaries and selects admissible restore points.
- ▪DART has been validated across multiple domains, showing improved recovery in commitment-sensitive scenarios.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.23311 (cs) [Submitted on 22 May 2026] Title:DART: Semantic Recoverability for Structured Tool Agents Authors:Ke Yang, Panpan Li, Zonghan Wu, Kejin Xu, Huaxi Huang, Xiaoshui Huang View a PDF of the paper titled DART: Semantic Recoverability for Structured Tool Agents, by Ke Yang and Panpan Li and Zonghan Wu and Kejin Xu and Huaxi Huang and Xiaoshui Huang View PDF HTML (experimental) Abstract:When a structured tool agent fails mid-execution, the runtime faces a dilemma: replaying the entire task is safe but wasteful, while restoring from a local checkpoint is efficient but can leave committed downstream work tied to an upstream history that no longer exists.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
