Scholarlink

The rapid growth of AI in mobile robotics highlights the need for scalable training methods that go beyond real-world data limitations. This paper presents the Digital–Synthetic–Real (DSR) Framework, which strategically integrates 70% simulated data (Isaac Sim), 20% synthetic data (Omniverse Cosmos), and 10% real-world data. The framework combines supervised learning (CNN + LSTM) with reinforcement learning, enabling quadruped robots to navigate urban environments using tactile surface indicators. By leveraging digital twins and annotated synthetic scenes, DSR reduces manual labeling, lowers training time by 30–40%, and achieves 92% accuracy with strong generalization to unseen conditions, these results demonstrate DSR as a practical and cost-effective methodology for accelerating AI development and deploying assistive robotics in real-world scenarios.

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

DSR Framework: A Hybrid Approach (Digital–Synthetic–Real) for Accelerated Training of AI Models in Mobile Robotics

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jhonattan Alexander Cordero Pareja

摘要

The rapid growth of AI in mobile robotics highlights the need for scalable training methods that go beyond real-world data limitations. This paper presents the Digital–Synthetic–Real (DSR) Framework, which strategically integrates 70% simulated data (Isaac Sim), 20% synthetic data (Omniverse Cosmos), and 10% real-world data. The framework combines supervised learning (CNN + LSTM) with reinforcement learning, enabling quadruped robots to navigate urban environments using tactile surface indicators. By leveraging digital twins and annotated synthetic scenes, DSR reduces manual labeling, lowers training time by 30–40%, and achieves 92% accuracy with strong generalization to unseen conditions, these results demonstrate DSR as a practical and cost-effective methodology for accelerating AI development and deploying assistive robotics in real-world scenarios.