AMR Task Allocation Using Deep Learning in Dynamic Warehouses
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for task allocation and path planning in autonomous mobile robots (AMRs) are inadequate, particularly in large-scale industrial environments, as they are complex, difficult to deploy, and unsuitable for heterogeneous systems, leading to sub-optimal and inefficient allocation of resources.
Innovation Solution
A Deep Learning-based neural network is implemented for AMR task allocation, which is model-free, scalable, and easy to deploy, allowing for continuous reassignment of tasks and handling of urgent scenarios, along with enhanced path-planning algorithms to optimize navigation and resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional task allocation algorithms are used for AMR systems, then task allocation can be performed, but the system complexity increases and deployment becomes difficult
Solution Approach 1:
The patent replaces complex mechanical/algorithms-based task allocation systems with a Deep Learning neural network model. The DL-based model learns optimal task allocation strategies from training data and provides predictions directly, eliminating the need for complex mathematical formulations and algorithm implementations that characterize conventional approaches.
Solution Approach 2:
The patent uses a trained Deep Learning model that captures the essence of complex task allocation algorithms in a simplified form. The model serves as a copy or approximation of the optimal allocation behavior, enabling easy deployment without implementing the underlying complex algorithms.
2Reliability
If conventional path planning algorithms are used for AMRs, then navigation can be achieved, but the allocation efficiency decreases in dynamic environments
Solution Approach 1:
The patent implements a dynamic task allocation system using Deep Learning that can adapt to changing environmental conditions in real-time. The model processes current system state inputs and generates updated task allocations dynamically, unlike static conventional algorithms that require re-computation when conditions change.
Solution Approach 2:
The patent performs preliminary training of the Deep Learning model offline using generated training datasets that represent various environmental scenarios. This preliminary action prepares the model to handle dynamic situations efficiently during deployment without requiring complex real-time computations.
3Adaptability or versatility
If heterogeneous AMR systems are deployed, then system versatility increases, but current allocation techniques become unsuitable
Solution Approach 1:
The patent develops a universal Deep Learning-based task allocation model that can handle heterogeneous AMR systems. The model learns from diverse training data representing different robot types, task categories, and environmental conditions, enabling it to generalize to heterogeneous configurations without requiring system-specific algorithm design.
4Adaptability or versatility
If complex allocation algorithms are implemented, then task allocation can be performed, but the deployment time increases
Solution Approach 1:
The patent performs extensive model training and validation in advance using generated training datasets. This preliminary action transfers the computational burden to the offline training phase, enabling rapid real-time predictions during deployment without requiring complex algorithm execution at runtime.
Data Source
AI summary
Techniques are disclosed to perform task allocation for autonomous systems by implementing machine-learning to perform task allocation to Autonomous Mobile Robots (AMRs) in an environment. The disclosed techniques also provide for enhanced path planning and the identification of AMR health and failure prediction to further improve upon task allocation and system efficiency.