• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2026, Vol. 62 ›› Issue (17): 120-135.doi: 10.3901/JME.260351

• 特邀专栏:人形机器人与具身智能 • 上一篇    

扫码分享

基于障碍物预测的轮式仿人机器人自适应混合路径规划方法

陶永1, 王潇桐2, 冯远宝1, 任帆3, 肖舒震1, 高赫2, 魏洪兴1   

  1. 1. 北京航空航天大学机械工程及自动化学院 北京 100191;
    2. 北京航空航天大学航空发动机研究院 北京 102206;
    3. 杭州市北京航空航天大学国际创新研究院(北京航空航天大学国际创新学院) 杭州 311115
  • 收稿日期:2025-12-04 修回日期:2026-03-13 发布日期:2026-11-20
  • 作者简介:陶永(通信作者),男,1979年出生,副教授,博士研究生导师。主要研究方向为智能机器人感知与决策、协作机器人力位控制、具身智能机器人等。E-mail:taoy@buaa.edu.cn
  • 基金资助:
    国家自然科学基金资助项目(52375005)。

Obstacle Prediction-based Adaptive Hybrid Path Planning Method for Mobile Dual-arm Robots

TAO Yong1, WANG Xiaotong2, FENG Yuanbao1, REN Fan3, XIAO Shuzhen1, GAO He2, WEI Hongxing1   

  1. 1. School of Mechanical Engineering and Automation, Beihang University, Beijing 100191;
    2. Research Institute of Aero-Engine, Beihang University, Beijing 102206;
    3. Hangzhou International Innovation Institute of Beihang University, Hangzhou 311115
  • Received:2025-12-04 Revised:2026-03-13 Published:2026-11-20

摘要: 针对轮式仿人机器人在复杂动态环境下路径规划中全局规划与局部避障协同不足、动态障碍物处理能力有限的问题,提出一种面向复杂动态环境的轮式仿人机器人自适应混合路径规划方法(Proactive dynamic window approach, PDWA)。该方法在统一的多目标优化框架下,融合改进的全局规划算法(Expanding disconnected graph star, EDG*)与局部动态窗口算法(Dynamic window approach, DWA),并结合动态障碍物检测与轨迹预测,实现路径规划对环境变化的实时响应。同时,引入预测编码方法构建“感知-预测-行动”闭环模型,用于调节规划过程中的权重分配,提升整体规划的连贯性与稳定性。最后,开展了轮式仿人机器人的仿真与实机实验。仿真结果表明,所提算法在复杂动态障碍环境中能够实现稳定且安全的导航,且在避障安全性、轨迹平滑性与速度稳定性方面均优于对比方法。机器人实机实验则进一步验证了所提方法在轮式仿人机器人与移动式协作机器人平台上的有效性,即在静态与动态障碍环境中均能生成平滑、连续的轨迹,并实现灵活、高效的动态避障,导航时间最短且运动安全可靠。研究结果表明,该方法在轮式仿人机器人路径规划场景中具有良好性能,为人形机器人在复杂多模态环境下的路径规划与运动控制提供了技术支撑。

关键词: 轮式仿人机器人, 路径规划, 混合算法, 运动预测, 自适应反馈控制, 动态避障

Abstract: To address the challenges faced by wheeled humanoid robots in complex dynamic environments—specifically, the insufficient coordination between global path planning and local obstacle avoidance, as well as the limited capability for handling dynamic obstacles—this study proposes an adaptive hybrid path planning method, the Proactive Dynamic Window Approach (PDWA). Under a unified multi-objective optimization framework, the method integrates an improved global planning algorithm, Expanding Disconnected Graph Star (EDG*), with the local Dynamic Window Approach (DWA), and incorporates dynamic obstacle detection and trajectory prediction to enable real-time responsiveness to environmental changes. In addition, a predictive coding mechanism is introduced to construct a “perception–prediction–action” closed-loop model that adaptively adjusts weight allocation during the planning process, enhancing the coherence and stability of overall planning. Simulation and real-world experiments were conducted on a wheeled humanoid robot. The simulation results show that the proposed method achieves stable and safe navigation in complex dynamic obstacle environments and outperforms comparison methods in obstacle avoidance safety, trajectory smoothness, and velocity stability. Real-robot experiments further validate the effectiveness of the proposed framework on both wheeled humanoid and mobile collaborative robot platforms, demonstrating its ability to generate smooth and continuous trajectories in static and dynamic environments while achieving flexible, efficient dynamic obstacle avoidance with the shortest navigation time and reliable motion safety. The results indicate that the proposed method exhibits strong performance in wheeled humanoid robot path planning scenarios and provides technical support for path planning and motion control of humanoid robots in complex multimodal environments.

Key words: wheeled humanoid robots, path planning, hybrid algorithm, motion prediction, adaptive feedback control, dynamic obstacle avoidance

中图分类号: