面向动态人群场景的机器人时空图分层导航

Hierarchical navigation of spatio-temporal graph for crowds

中文关键词  地面移动机器人  动态路径规划  滚动时域优化  自主导航

英文关键词  service robots  dynamic path planning  receding horizon control  crowd navigahtion

中文摘要

      人群场景是服务机器人的重要应用场景之一.然而,由于行人具有自主决策能力,并且行人行为具有不确 定性,如何实现人群场景下的自主安全导航,成为了当前服务机器人领域的一大挑战性问题.为此,本文提出了一种 面向动态人群场景的时空图的分层导航框架(STG-HCN).首先,本文引入德劳内三角剖分描述动态人群拓扑关系, 并拓展到时间维度,进而提出一种时空图搜索算法,实现全局路径的快速生成.其次,考虑到行人动态属性,本文进 一步引入行人的速度和视角信息,提出一种基于高斯分布的前向注意力势场,用于评估全局路径对行人的影响.而 后, 针对动态行人构建时变避碰约束,并且引入滚动时域优化实现时变约束优化问题的求解,进一步提升机器人全 局路径跟踪的运动安全.最后,为了验证算法的有效性,本文开展了大量的仿真及实物实验.仿真实验结果表明, STG-HCN可将现有基准算法的平均碰撞次数降低55.7%,而基于Fetch机器人的实物实验充分说明了算法在实际人 群场景中的有效性.

英文摘要

      Crowd environment is one of the important application scenarios for ground mobile robots. However, due to pedestrians’ decision-making abilities and the inherent uncertainty of their behavior, achieving autonomous and safe navigation in crowded environments has emerged as a significant challenge in the field of service robots. To address this, this paper proposes a spatio-temporal-graph-based hierarchical crowd navigation (STG-HCN) that provides both global and local solutions. Firstly, the STG-HCN introduces Delaunay triangulation to describe the topological relationship of crowd s. This approach gives rise to a spatio-temporal graph search algorithm, facilitating the rapid generation of global paths. Secondly, considering the dynamic attributes of pedestrians, the paper further integrates pedestrian speed and perspective information. This leads to the proposal of a forward attention potential field based on a Gaussian distribution, aimed at evaluating the impact of the global path on pedestrians. Furthermore, time-varying collision avoidance constraints are for mulated for dynamic pedestrians, and a receding horizon optimization framework is introduced to solve the time-varying constrained optimization problem, thereby significantly enhancing the motion safety of the robot’s global path tracking. To validate the effectiveness of the proposed algorithm, this paper conducts extensive simulation and real-world experiments. Simulation results indicate that STG-HCN can reduce the average number of collisions by 55.7% compared to the bench mark algorithms. Additionally, real-world experiments conducted using a Fetch robot demonstrate the algorithm’s efficacy in practical crowd scenes.