Abstract:Robot path planning in 3D space is a challenging problem for its complex configuration. Sampling-based algorithms have gained great success in solving path planning problems in 3D space, but the quality of the initial path is not guaranteed and the convergence to the optimal solution is slow. To address these problems, in this article, we present a novel sampling-based path planning framework enhanced by the deep neural network (DNN) with applications to 3D space. In the proposed framework, we first train the DNN with a number of successful path planning cases in 3D space. Then the DNN is utilized to predict the promising region where the feasible path probably exists for a given path planning problem. This predicted promising region serves as a nonuniform sampling heuristic to bias the sampling process of the path planner. In this way, the path planner can focus on the promising region in the exploration and exploitation process so that the path planning speed gets accelerated. We conduct numerical simulations to evaluate the performance of the proposed algorithm and the results show that it can perform much better than conventional path planning algorithms. Furthermore, we also investigate the performance of different DNN architectures for path planning in 3D space. Note to Practitioners—In this work, we aim to provide an efficient learning-based method to accelerate the robot path planning process in 3D space. Conventional path planning algorithms need to perceive the environment first, and then implement a series of calculations such as collision checking and data storing to generate a feasible path. When facing complex and high-dimensional environments, they do not perform well. But the proposed neural network method in this article can predict the promising region where the feasible path exists for any given environment. This prediction result is used to guide the path planning process so that the algorithm performance can get significantly improved. Apart from sampling-based algorithms, the proposed neural network model can also be extended to other types of path planning algorithms.

Predicting Sample Collision with Neural Networks

Neural Collision Clearance Estimator for Batched Motion Planning

Safe and Efficient Path Planning under Uncertainty via Deep Collision Probability Fields

Neural Implicit Swept Volume Models for Fast Collision Detection

Reducing Collision Checking for Sampling-Based Motion Planning Using Graph Neural Networks

Deeply Informed Neural Sampling for Robot Motion Planning

Speeding Up Optimization-based Motion Planning through Deep Learning

A Short-term Motion Prediction Approach for Guaranteed Collision-Free Planning.

Direct Robot Configuration Space Construction using Convolutional Encoder-Decoders

Learning-Based Proxy Collision Detection for Robot Motion Planning Applications

DeepCollide: Scalable Data-Driven High DoF Configuration Space Modeling using Implicit Neural Representations

Collision-Affording Point Trees: SIMD-Amenable Nearest Neighbors for Fast Collision Checking

Deep Neural Network Enhanced Sampling-Based Path Planning in 3D Space

Learning Sampling Distribution for Motion Planning with Local Reconstruction-Based Self-Organizing Incremental Neural Network

CabiNet: Scaling Neural Collision Detection for Object Rearrangement with Procedural Scene Generation

Configuration Space Decomposition for Learning-based Collision Checking in High-DOF Robots

Robot Active Neural Sensing and Planning in Unknown Cluttered Environments

Multi-Robot Guided Sampling-Based Motion Planning With Dynamics in Partially Mapped Environments

Semantically-enhanced Deep Collision Prediction for Autonomous Navigation using Aerial Robots

An Universal GMMs-based Sample Strategy and Collision Checker for Robot Motion Planning

Graph Wasserstein Autoencoder-Based Asymptotically Optimal Motion Planning with Kinematic Constraints for Robotic Manipulation.