Neural Network Task Planning for Dynamic Search Trees
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating task plans for autonomous systems, such as robots and vehicles, rely on symbolic automated planning, which can be inefficient and limited in handling complex tasks, as they express information using symbols and logical operations, making it difficult to adapt to dynamic environments.
Innovation Solution
A method and apparatus that utilize a neural network to generate task plans by creating a search tree based on task states and actions, estimating a recommended path through the network, and determining a target path from an initial to a target state, incorporating techniques like one-hot encoding and hash operations to process sequence data and generate training data for the neural network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If symbolic automated planning technology is used to generate task plans, then the system can perform logical operations on task information, but the system becomes inefficient and limited when handling complex tasks in dynamic environments
Solution Approach 1:
The patent replaces the mechanical symbolic logical operation system with a neural network-based system. The neural network learns patterns from training data consisting of task states, actions, and outcomes, enabling the system to adapt to dynamic environments through learned representations rather than rigid symbolic rules. This substitution allows the system to handle complex tasks more efficiently by leveraging the neural network's ability to generalize from training examples.
2Productivity
If symbolic logical operations are used to process task information, then the system can reason about tasks, but the system complexity increases and performance decreases for complex tasks
Solution Approach 1:
The patent substitutes complex symbolic logical operations with a neural network that processes task information through learned patterns. The neural network takes task states and actions as input and directly predicts optimal actions or task outcomes, eliminating the need for complex symbolic reasoning mechanisms. This reduces system complexity while improving performance on complex tasks.
Solution Approach 2:
The patent transforms discrete symbolic task representations into continuous neural network activations. By encoding task states, actions, and outcomes as numerical vectors that can be processed by the neural network, the system changes the parameter space from discrete symbols to continuous values, enabling more efficient processing of complex tasks.
3Adaptability or versatility
If traditional search methods are used to find task plans, then the system can explore possible actions, but the resource consumption increases and the system cannot efficiently handle complex task spaces
Solution Approach 1:
The patent performs preliminary training of the neural network on a large dataset of task examples before actual task execution. This preliminary action allows the neural network to learn optimal strategies and patterns in advance, so that during actual task execution, the system can quickly generate task plans without extensive real-time searching. This significantly reduces resource consumption during operation.
Solution Approach 2:
The patent uses training data that copies successful task execution patterns from expert demonstrations or pre-computed solutions. By learning from these copied examples, the neural network internalizes effective strategies, allowing the system to handle complex tasks efficiently without exhaustively searching the task space during execution.
Data Source
AI summary
Provided are a method and an apparatus for generating a task plan. A method of generating a task plan for performing an arbitrary task includes generating a search tree based on a plurality of task states of the task and a plurality of task actions for performing the task, estimating a recommended path for an internal connection of the search tree by inputting the plurality of task states and the plurality of task actions to a neural network based on the search tree, and generating the task plan by determining a target path that reaches from an initial state of the task to a target state of the task based on the recommended path.