Beyond Global Replanning: Hierarchical Recovery for Cross-Device Agent Systems

2026-06-18 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Advanced, quick

Summary

H-RePlan is a hierarchical replanning framework designed for multi-device agent systems that operate across diverse applications and devices, addressing dynamic runtime failures. Existing systems often rely on coarse-grained recovery, retrying strategies or revising global plans without modeling device-local strategy space. H-RePlan overcomes this by equipping each device with interchangeable execution strategies and separating device-local strategy recovery from orchestrator-level global replanning via a compact cross-layer failure abstraction. Evaluated using HeraBench, a fault-injected benchmark for Linux and Android workflows, H-RePlan significantly outperforms single-strategy and coarse-grained baselines. It achieves higher completion, instruction adherence, and perfect-pass rates, while also reducing the token cost for reliable end-to-end success, demonstrating the necessity of scope-aware hierarchical recovery.

Key takeaway

For AI Engineers building or deploying multi-device agent systems, recognize that current coarse-grained recovery methods are insufficient for dynamic real-world environments. You should integrate hierarchical recovery frameworks like H-RePlan, which separate device-local strategy handling from global replanning. This approach will significantly improve your system's robustness, completion rates, and resource efficiency, reducing token costs for reliable task execution across heterogeneous devices.

Key insights

Hierarchical, scope-aware recovery is crucial for robust multi-device agent systems operating under dynamic runtime failures.

Principles

• Decouple device-local strategy recovery from global replanning.
• Employ interchangeable execution strategies at the device level.
• Utilize compact cross-layer failure abstraction for efficiency.

Method

H-RePlan equips each device with interchangeable execution strategies and separates device-local strategy recovery from orchestrator-level global replanning using a compact cross-layer failure abstraction to manage failures.

In practice

• Implement device-specific recovery strategies for agent robustness.
• Design clear failure abstraction layers between device and orchestrator.
• Evaluate multi-device agents using fault-injected benchmarks.

Topics

• Multi-device Agents
• Hierarchical Replanning
• Fault Recovery
• Agent Orchestration
• Cross-Device Workflows
• HeraBench Benchmark

Best for: Research Scientist, AI Architect, AI Scientist, AI Engineer, Software Engineer

Related on AIssential

• Decomposing the Agent Orchestration System: Lessons Learned — Agents can reliably recover from infrastructure failures if given the right context and capabilities.
• LAI #115: The Hidden Cost of “Agent-First” Thinking — AI system failures often stem from architectural and reproducibility issues, not just model performance.
• Discord Reveals How a Hidden Circular Dependency Triggered Its March Voice Outage — Hidden circular dependencies can trigger cascading failures in highly resilient distributed systems.
• System Design for Maintaining Internal State Consistency in Long-Horizon Robotic Tabletop Games — Explicit state partitioning and recovery mechanisms are critical for robotic system reliability in long-horizon, turn-based interactions.
• LLM Fallbacks Break Agent Pipelines — I Built the Missing Recovery Layer — LLM agent pipeline fallbacks must adapt payloads and preserve state to maintain data schema integrity, not just ensure completion.
• Building Self-Healing Software Systems with Multi-Agent AI Architectures — Multi-agent AI enables autonomous detection and resolution of software system issues.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.