📄 中文摘要
提出了一种新的神经采样器GAIDE，旨在解决基于采样的运动规划算法在高维配置空间中的样本效率问题。传统的采样方法通常依赖于均匀或手工设计的采样原语，导致在复杂环境中的规划效率低下。GAIDE通过学习先前规划经验中的采样分布，能够更有效地引导运动规划器朝向规划目标。此外，GAIDE利用空间结构信息，增强了对运动规划问题固有空间结构的编码能力，从而提升了运动规划的性能和效率。该方法在多个实验中展示了其在复杂场景下的有效性和优势。
GAIDE：基于图的注意力掩蔽用于空间和体现感知的运动规划
📄 English Summary
GAIDE: Graph-based Attention Masking for Spatial- and Embodiment-aware Motion Planning
A novel neural informed sampler, GAIDE, is proposed to address the sample inefficiency of sampling-based motion planning algorithms in high-dimensional configuration spaces. Traditional sampling methods often rely on uniform or handcrafted sampling primitives, leading to low planning efficiency in complex environments. GAIDE learns the sampling distribution from prior planning experiences, effectively guiding the motion planner towards the planning goal. Additionally, it leverages spatial structure information to enhance the encoding of the inherent spatial structures in motion planning problems, thereby improving the performance and efficiency of motion planning. The method demonstrates its effectiveness and advantages in complex scenarios through various experiments.