AI Agent Planning Module for Dynamic Tool Invocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI agents face challenges such as coarse or fine module granularity in their architecture, reliance on manually set rules, incomplete function implementation, lack of advanced cognitive capabilities like reflection, and inefficient memory usage.
Innovation Solution
The proposed solution involves an artificial intelligence-based information processing method and apparatus that includes determining execution information for processing input data, retrieving or invoking necessary memory or tool information, and integrating processing results to generate output feedback. This approach enhances the AI agent's planning, action, evaluation, and reflection capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If AI agents use coarse or fine module granularity in their architecture, then the system structure becomes more organized, but the complexity of system design and implementation increases
Solution Approach 1:
The patent segments the AI agent architecture into distinct functional modules including a large language model module, a planning module, an action module, and a memory module. Each module has specific responsibilities: the planning module generates plans based on user input, the action module executes tools and operations, and the memory module stores and retrieves information. This segmentation allows for organized system structure while managing complexity through clear module boundaries and interfaces.
2Ease of operation
If AI agents rely on manually set rules, then the system is easier to control, but the adaptability to unknown environments decreases
Solution Approach 1:
The patent implements dynamic adaptability through the planning module that can generate and adjust plans based on real-time feedback from action execution results. The system dynamically selects which tools to invoke, what memory to retrieve, and how to integrate results, rather than following fixed manual rules. This allows the AI agent to adapt to unknown environments while maintaining controllability through the structured planning and execution framework.
Solution Approach 2:
The patent incorporates feedback mechanisms where the execution results of actions are fed back to the planning module. The planning module uses this feedback to evaluate whether the current plan is effective and to generate subsequent plans or adjustments. This feedback loop enables the system to adapt to unknown environments by learning from execution outcomes while maintaining control through the structured planning-process-evaluation cycle.
3Productivity
If AI agents implement incomplete function implementation, then the development speed increases, but the reliability of task completion decreases
Solution Approach 1:
The patent implements preliminary action through the planning module that generates comprehensive plans before execution. The planning module anticipates required actions, tool invocations, and memory retrievals in advance, creating a structured execution roadmap. This preliminary planning ensures that even with incomplete implementation of individual functions, the overall task completion reliability is maintained through proper sequencing and coordination of available functions.
Solution Approach 2:
The patent implements a unified planning module that serves multiple functions: generating initial plans, selecting tools to invoke, determining memory retrieval needs, and evaluating execution results. This multi-functional planning module provides a universal framework that can coordinate incomplete implementations of various specific functions while maintaining overall task completion reliability through centralized planning and coordination.
4Device complexity
If AI agents lack advanced cognitive capabilities like reflection, then the system complexity is reduced, but the intelligence and problem-solving ability decrease
Solution Approach 1:
The patent introduces the planning module as an intermediary between the large language model and the action execution system. This planning module provides reflection-like capabilities by evaluating execution results, determining whether plans are on track, and generating subsequent plans or adjustments. The planning module acts as a mediator that adds cognitive evaluation and reasoning capabilities without requiring complex changes to the underlying large language model or action execution mechanisms.
5Use of energy by moving object
If AI agents use inefficient memory usage, then the resource consumption is reduced, but the processing speed and accuracy of information retrieval decrease
Solution Approach 1:
The patent implements local quality in memory usage by having the planning module selectively determine what memory to retrieve based on the specific task and context. Rather than uniformly accessing all memory, the system retrieves only relevant memory portions identified by the planning module's analysis of the current situation and plan requirements. This selective memory retrieval optimizes both resource consumption and retrieval efficiency by focusing computational resources on relevant information.
Data Source
AI summary
A method for processing information is provided. The method includes obtaining input information to be processed. The method further includes determining execution information associated with processing of the input information. The execution information includes at least one of memory information to be retrieved or tool information to be invoked. The method further includes obtaining, by using the execution information, at least one piece of processing result information corresponding to the processing of the input information. The method further includes the at least one piece of processing result information to generate output information for feedback.