SIMPNet: Spatial-informed Motion Planning Network

Davood Soleymanzadeh1, Xiao Liang2, Minghui Zheng1.
1 J. Mike Walker '66 Department of Mechanical Engineering, Texas A&M University, 2 Zachry Department of Civil and Environmental Engineering, Texas A&M University.
Paper arXiv Code

Abstract

Current robotic manipulators require fast and efficient motion-planning algorithms to operate in cluttered environments. State-of-the-art sampling-based motion planners struggle to scale to high-dimensional configuration spaces and are inefficient in complex environments. This inefficiency arises because these planners utilize either uniform or hand-crafted sampling heuristics within the configuration space. To address these challenges, we present the Spatial-informed Motion Planning Network (SIMPNet). SIMPNet consists of a stochastic graph neural network (GNN)-based sampling heuristic for informed sampling within the configuration space. The sampling heuristic of SIMPNet encodes the workspace embedding into the configuration space through a cross-attention mechanism. It encodes the manipulator's kinematic structure into a graph, which is used to generate informed samples within the framework of sampling-based motion planning algorithms. We have evaluated the performance of SIMPNet using a UR5e robotic manipulator operating within simple and complex workspaces, comparing it against baseline state-of-the-art motion planners. The evaluation results show the effectiveness and advantages of the proposed planner compared to the baseline planners.

Video Introduction

A brief video introduction of SIMPNet.

Method

SIMPNet's Sampling Heuristic Strucutre

We use graph neural networks (GNNs) to incorporate the structure of the robotic manipulator's kinematic chain into our proposed sampling heuristic. This approach makes our planner aware of the manipulator's articulated structure. We also use an attention mechanism to condition the planning problem on workspace embeddings, given different dimensions of configuration space and workspace in robotic manipulators.

Interpolate start reference image.

Schematic of spatial-informed sampling heuristic within SIMPNet.


Robotic Manipulator's Graph Representation

We build a graph that implicitly mimics the kinematic chain of the robotic manipulator.

Interpolate start reference image.

Results

Workspaces

Type of workspaces used for training and evaluation of SIMPNet (i.e., simple and complex).

Interpolate start reference image.

Planning Results vs. Benchmarks

Time vs. success rate and vs. planning cost of SIMPNet compared to benchmark planners.

Interpolate start reference image.

Real-world Deployment

SIMPNet versus Benchmark Planners

Experimental evaluation and comparison of planned paths by SIMPNet and some baseline planners in a complex environment: Bi-RRT (Bi-directional RRT), IRRT* (Informed-RRT*), and MPNet (Motion Planning Networks). RRT completes the path in 32 seconds with a planning cost of 9.32. Bi-RRT plans in 9.44 seconds with a planning cost of 9.02. IRRT* takes 500 seconds, achieving a cost of 4.01 MPNet completes the path in 218 seconds with a planning cost of 11.3. SIMPNet plans the path in 5.81 seconds with a planning cost of 8.08.

Interpolate start reference image.

Comparison of real-time deployment of SIMPNet and benchmark planners.


Planning Videos

Bi-RRT (x8)

IRRT* (x4)

MPNet (x4)

SIMPNet (x4)

BibTeX

@article{soleymanzadeh2025simpnet,,
  title     = {SIMPNet: Spatial-Informed Motion Planning Network},
  author    = {Soleymanzadeh, Davood and Liang, Xiao and Zheng, Minghui},
  journal   = {IEEE Robotics and Automation Letters},
  year      = {2025},
  publisher = {IEEE}
}