Scholarlink

<p>Collision monitoring for Eye-in-Hand dual-arm robots is constrained by strict real-time budgets, making exhaustive high-fidelity checking across all component pairs impractical. This challenge is particularly acute for fragile heterogeneous end-effectors such as wrist-mounted cameras. To address it, we propose a Semantics-Enhanced Dynamic Hierarchical Filtering (SDHF) framework for fixed-budget collision monitoring. SDHF combines motion-trend-guided spatial pre-screening, dual-threshold candidate retention, and exact heterogeneous geometric evaluation using capsules for links and oriented bounding boxes (OBBs) for cameras. Experiments include parameter calibration on 1,200 extreme-proximity samples, runtime and tail-latency validation on progressive scenarios and DualArm-100&#xa0;K, and physical safety-function evaluation on General-500. Under the tested conditions, SDHF achieves an average per-frame algorithmic runtime of 7.8 ms and a P99 algorithmic runtime below 10 ms in the runtime benchmarks. On General-500, it reaches 96.40% precision and a 98.17% F1-score, while no missed detections were observed on the evaluated set within the tested range. In the representative physical loop, the measured average and P99 end-to-end latencies are 9.995 ms and 19.953 ms, respectively.</p>

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

SDHF for real-time collision monitoring in Eye-in-Hand dual-arm robots

  • 期刊论文

  • 发表日期：

    2026年7月9日

查看全文
  • Kunning Ji,
  • Chunyang Liu,
  • Yan Huang,
  • Xin Sui,
  • Nan Guo,
  • Liuzhen Wang,
  • Xin Chen,
  • Tian Gao

摘要

Collision monitoring for Eye-in-Hand dual-arm robots is constrained by strict real-time budgets, making exhaustive high-fidelity checking across all component pairs impractical. This challenge is particularly acute for fragile heterogeneous end-effectors such as wrist-mounted cameras. To address it, we propose a Semantics-Enhanced Dynamic Hierarchical Filtering (SDHF) framework for fixed-budget collision monitoring. SDHF combines motion-trend-guided spatial pre-screening, dual-threshold candidate retention, and exact heterogeneous geometric evaluation using capsules for links and oriented bounding boxes (OBBs) for cameras. Experiments include parameter calibration on 1,200 extreme-proximity samples, runtime and tail-latency validation on progressive scenarios and DualArm-100 K, and physical safety-function evaluation on General-500. Under the tested conditions, SDHF achieves an average per-frame algorithmic runtime of 7.8 ms and a P99 algorithmic runtime below 10 ms in the runtime benchmarks. On General-500, it reaches 96.40% precision and a 98.17% F1-score, while no missed detections were observed on the evaluated set within the tested range. In the representative physical loop, the measured average and P99 end-to-end latencies are 9.995 ms and 19.953 ms, respectively.