End-to-End Path Planning Under Linear Temporal Logic Specifications

End-to-End Path Planning Under Linear Temporal Logic Specifications

This paper presents a novel deep learning framework for robotic path planning that seamlessly integrates Linear Temporal Logic (LTL) with trajectory optimization to meet mission specifications efficiently. Our approach innovates on several fronts: First, by training a neural network end-to-end to ge...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access Vol. 12; pp. 57410 - 57423
Main Authors: Yang, Chaeeun, Yoon, Sojeong, Cho, Kyunghoon
Format: Journal Article
Language:English
Published: Piscataway IEEE 2024
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects:
Deep learning
Deep learning-based control synthesis
formal methods
Formal specifications
Machine learning
Mission critical systems
mission-based path planning
Navigation
Neural networks
Optimal control
Path planning
Probabilistic models
Robots
Specifications
Task analysis
Temporal logic
Trajectory
Trajectory optimization
Uncertainty
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper presents a novel deep learning framework for robotic path planning that seamlessly integrates Linear Temporal Logic (LTL) with trajectory optimization to meet mission specifications efficiently. Our approach innovates on several fronts: First, by training a neural network end-to-end to generate control sequences that are not only cost-effective but also fully compliant with LTL-defined mission objectives. This negates the need for generating traditional automatons, as our network is capable of directly interpreting LTL formulas to guide path planning. Key to our framework is the use of a Conditional Variational Autoencoder (CVAE), which is adept at identifying the optimal distribution of trajectories. This enables the extraction of practical control sequences through a process of sampling latent variables and inferring control outputs, thus addressing the critical challenge of trajectory optimization under uncertainty. Moreover, our model incorporates transformer networks to refine these trajectory distributions into nearly optimal control sequences, further enhanced by a Gaussian Mixture Model (GMM) to manage uncertainty and fine-tune adjustments effectively. Empirical validation through comparative simulations showcases the superior performance of our model. It achieves significant advancements in trajectory optimality and mission success rates over existing deep learning-based path planning strategies. This work underscores the potential of integrating LTL in deep learning models for robotic path planning, marking a significant leap forward in the domain.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2024.3392289