SafeOrchestrate: Runtime-verified, intent-driven orchestration for cyber-physical systems
SafeOrchestrate: Runtime Checked Intent-Driven Orchestration for Cyber-Physical Systems
AI summary
The researchers propose SafeOrchestrate, a runtime-verified, intent-driven orchestration architecture for cyber-physical systems. An LLM generates LTL specifications from natural language intents, connects to device Thing Descriptions via MCP, and verifies each candidate action with an SMT-based guardrail before execution. In simulated tasks for smart homes, greenhouses, and industrial automation, its required-action accuracy was 100% in all cases, and MCP execution similarity reached 100% except for one item at 92%, whereas the baseline of an LLM directly connected to MCP dropped to 13% on multi-device, event-triggered, and ordered tasks.