September 25, 2024
In this paper, we propose an intelligent planning and control system for autonomous vehicles
operating in high-dimensional spaces. Our solution is a fully unsupervised scheduler and motion
planner. Many warehouses utilize automated material handling processes to expedite product
transshipment; however, as the dimensionality of the space grows, the complexity of the control
system increases significantly.
The proposed model leverages a kernel-based control system that employs Deep Reinforcement
Learning (DRL) to manage low-dimensional spaces. Additionally, it integrates a global
transition-control system to effectively coordinate communication between kernels. This global
system generates virtual paths for each product, assigns tasks to specific kernels for handling
products within their zones, and ensures smooth transitions between different blocks to guide
each product to its destination.
Our approach demonstrates strong performance in terms of both speed and the number of
movements required. Furthermore, the system is robust to increases in both warehouse size and
the number of products.
For more details :
View article document
