Scholarlink

This project explored the feasibility and ethical implications of designing a human digital twin (HDT) to emulate and optimize human operator cognition in high-demand, high-risk battle management command, control, and communications (C3) environments. We first performed a qualitative meta-analysis and cognitive risk analysis of the significant cognitive challenges of the C3 operator. We then designed a preliminary HDT concept, using a situational awareness-oriented design (SAOD) approach and human-centered reinforcement learning (HCRL) techniques, intended to be implemented independently or teamed with a human operator to assist with functions that require high cognitive demand. To achieve this, we plan on implementing an RL model architecture to serve as an HDT. Our agent will first undergo supervised learning on data from C3 subject matter experts (SMEs) engaging with simulated air C3 scenarios. Data collected from the simulated battlespace and SME feedback will be used to train the agent to inform our RL reward functions and human-in-the loop (HITL) training. After baseline models are developed, they will be fine-tuned through self-play to meet or exceed functional criteria as defined in our C3 operational context. This research will offer valuable insights into the future of DT technology and its practical applications for decision-support systems in operational environments.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of Human Centered Reinforcement Learning to Develop an Air Battle Management Operator Digital Twin

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Courtney L. Crooks,
  • Jasmine Davidson,
  • Vincent Davidson,
  • Patrick Tyne,
  • Katy Wills,
  • Peter Lebedev,
  • Tyler Rowe,
  • Eric Tang,
  • Jessica Inman

摘要

This project explored the feasibility and ethical implications of designing a human digital twin (HDT) to emulate and optimize human operator cognition in high-demand, high-risk battle management command, control, and communications (C3) environments. We first performed a qualitative meta-analysis and cognitive risk analysis of the significant cognitive challenges of the C3 operator. We then designed a preliminary HDT concept, using a situational awareness-oriented design (SAOD) approach and human-centered reinforcement learning (HCRL) techniques, intended to be implemented independently or teamed with a human operator to assist with functions that require high cognitive demand. To achieve this, we plan on implementing an RL model architecture to serve as an HDT. Our agent will first undergo supervised learning on data from C3 subject matter experts (SMEs) engaging with simulated air C3 scenarios. Data collected from the simulated battlespace and SME feedback will be used to train the agent to inform our RL reward functions and human-in-the loop (HITL) training. After baseline models are developed, they will be fine-tuned through self-play to meet or exceed functional criteria as defined in our C3 operational context. This research will offer valuable insights into the future of DT technology and its practical applications for decision-support systems in operational environments.