Agentic AI-powered human-in-the-loop cyber-physical systems with robustness and determinism guarantees. Foundation model integration for mission-critical CPS.
本文研究了基于大语言模型(LLM)的智能体在关键任务人机混合网络物理系统(CPS)中的应用,提出了Agentic Driving Coach框架,解决智能体AI在CPS中的鲁棒性和确定性挑战。
class AgenticCPS:
def __init__(self, foundation_model, safety_monitor):
self.llm = foundation_model
self.safety = safety_monitor
self.state = SystemState()
def decision_loop(self, observations):
# 智能体推理
intent = self.llm.reason(observations, self.context)
# 安全检查
if not self.safety.validate(intent):
intent = self.safety.fallback_action()
# 确定性执行
return self.deterministic_execute(intent)
┌─────────────────┐
│ LLM Agent │ 意图生成
├─────────────────┤
│ Safety Monitor │ 安全验证
├─────────────────┤
│ Deterministic │ 确定性执行
│ Controller │
├─────────────────┤
│ Physical System │ 物理层
└─────────────────┘