Speaker
Description
As laser-driven plasma accelerator facilities grow in complexity, operators face increasing challenges in cross-subsystem control, historical data correlation and rapid troubleshooting. At Peking University’s Compact Laser Plasma Accelerator II (CLAPA-II), we have deployed an intelligent control assistant built upon Osprey—an agentic interface and safety harness framework designed for safety-critical control systems. Our implementation integrates the AI agent into the CLAPA-II control environment across three core dimensions. First, it establishes direct integration with the EPICS control stack, translating natural language operator intents into auditable channel access operations under strict human-in-the-loop safety constraints. Second, the agent connects to the EPICS Archiver Appliance to autonomously retrieve, visualize, and analyze historical PV time-series data for anomaly detection and post-shot diagnostics. Third, we built a specialized knowledge base for the CLAPA-II trigger system to provide the agent with comprehensive foundational knowledge of our timing infrastructure. Applying Osprey to the trigger system effectively bridges the knowledge gap in specialized experimental scenarios, enabling intelligent control and monitoring while assisting operators with routine maintenance, commissioning, and system upgrades. This talk presents the architecture, operational workflows, and initial practical experiences of deploying Osprey at CLAPA-II.
