CN-122021835-A - Community service robot intelligent decision-making system based on large model and behavior tree

CN122021835ACN 122021835 ACN122021835 ACN 122021835ACN-122021835-A

Abstract

The invention belongs to the technical field of service robots and artificial intelligence, discloses an intelligent decision-making system of a community service robot based on a large model and a behavior tree, and solves the problems that a traditional community service robot is weak in complex instruction understanding, stiff in decision logic, unreliable in execution and free of autonomous learning ability. The system comprises an instruction input module, an intention analysis and task arrangement module, a behavior tree construction module, a behavior execution module, an autonomous learning module and a knowledge base module, wherein complex instructions are cooperatively disassembled into a structured sub-action sequence through a cloud edge model, the behavior tree is automatically constructed and executed based on an ROS system, the autonomous learning is realized by depending on a double-agent evaluation mechanism, and the knowledge base stores community scene and robot capability data. The intelligent and service reliability of the robot in the intelligent community is remarkably improved, and the intelligent and service reliability intelligent robot is suitable for scenes such as reception guidance, article delivery, safety inspection and the like.

Inventors

QING LINBO
KUANG HAOZE
YANG HONG
WANG PINGYU
CHEN HONGGANG

Assignees

天府绛溪实验室

Dates

Publication Date: 20260512
Application Date: 20260409

Claims (6)

1. A community service robot intelligent decision system based on a large model and a behavior tree is characterized by comprising: (1) The instruction input module is used for receiving and preprocessing voice or text service instructions input by a user; (2) The intention analysis and task arrangement module is connected with the instruction input module and is used for judging the instruction type, calling a large language model and combining a knowledge base and an example base when confirming the control intention, and analyzing and disassembling the complex instruction into a structured sub-action sequence; (3) The behavior tree construction module is connected with the intention analysis and task arrangement module and is used for analyzing the sub-action sequences and automatically constructing a behavior tree consisting of a control node, an execution node and a decoration node; (4) The behavior execution module is connected with the behavior tree construction module and is used for dispatching and executing the behavior tree so as to control the robot to complete corresponding actions; (5) The autonomous learning module is connected with the intention analysis and task arrangement module and the behavior execution module and is used for collecting task execution samples and automatically expanding the effective samples to the example library through an evaluation mechanism; (6) And the knowledge base module is connected with the intention analysis and task arrangement module and is used for storing community scene environment information and robot hardware capability information.
2. The community service robot intelligent decision system based on the large model and the behavior tree according to claim 1 is characterized in that the intent analysis and task arrangement module comprises an edge lightweight model unit, a cloud large language model unit, a behavior library, an example library and a similar example retrieval unit, wherein the edge lightweight model unit is used for carrying out quick intent classification on input instructions, the cloud large language model unit is used for carrying out deep analysis and task planning on control instructions, the behavior library is used for storing basic actions executable by a robot and parameter interfaces thereof, the example library is used for storing the corresponding relation between the instructions and sub-action sequences which are verified to be correct, and the similar example retrieval unit is used for matching historical similar examples for current instructions through text vectorization and vector database retrieval.
3. The community service robot intelligent decision system based on the large model and the behavior tree according to claim 1, wherein the behavior tree construction module converts a sub-action sequence in the JSON format into the behavior tree by adopting a recursive degradation analysis algorithm, and utilizes a 'behavior tree blackboard' as a global shared data area to realize parameter transmission and state sharing among nodes.
4. The intelligent decision-making system of the community service robot based on the large model and the behavior tree according to claim 1, wherein the behavior execution module adopts a multi-process architecture to run the behavior tree executor and comprises an execution state monitoring and node retry mechanism.
5. The community service robot intelligent decision system based on the large model and the behavior tree according to claim 1, wherein the autonomous learning module comprises a behavior recorder, a sample diversity evaluation agent and a sample correctness evaluation agent, wherein the behavior recorder is used for recording user instructions, generated action sequences and execution results, the sample diversity evaluation agent is used for judging uniqueness of a new sample based on vector similarity, and the sample correctness evaluation agent is used for calling a plurality of large language models to evaluate action sequence logic of the new sample.
6. The intelligent decision-making system of the community service robot based on the large model and the behavior tree according to claim 1, wherein the information stored by the knowledge base module comprises the key point positions of the community indoor map, the IDs of access control equipment, the channel attributes and the performance parameters of the robot movement and interaction equipment.

Description

Community service robot intelligent decision-making system based on large model and behavior tree Technical Field The invention relates to the technical field of service robots and artificial intelligent decision making, in particular to a community service robot intelligent decision making system integrating a Large Language Model (LLM) and a Behavior Tree (BT), which is suitable for scenes requiring understanding of complex instructions and reliable execution, such as reception guidance, article delivery, safety inspection and the like in an intelligent community. Background With the advancement of intelligent community construction, service robots have become an important tool for improving community management efficiency and resident life quality. However, the intelligent decision system of the existing community service robot has obvious limitations in practical application: First, complex instructions are weak in resolving power. Conventional systems rely heavily on a predefined set of simple instructions that cannot understand complex task instructions issued in natural language by users that contain multi-step and fuzzy semantics. This results in an insufficient level of robot intelligence and poor user experience. Secondly, decision logic is stiff and has poor expansibility. Most systems employ decision logic based on fixed rules or finite state machines, requiring a developer to manually write code for each new task. When community service requirements change or task types are newly added, the system is difficult to quickly adapt, and the maintenance and updating costs are high. Again, task execution reliability is inadequate. A complex task is typically made up of multiple sub-actions with dependencies. The existing system lacks of standardized description and unified scheduling of execution logic, and is easy to interrupt or fail task execution due to action sequence errors, state management confusion or abnormal processing missing. Finally, autonomous learning and optimization capabilities are lacking. The system cannot learn from historical execution experience and decision capability is statically cured. In the face of new scenes and new instructions which are continuously emerging in communities, the intelligent community dynamic development system is completely dependent on manual upgrading, cannot realize autonomous evolution, and is difficult to meet the requirements of the intelligent community dynamic development. Therefore, what is needed is a community service robot intelligent decision system that can understand complex instructions, has flexible and reliable execution logic, and can be optimized continuously through autonomous learning. Disclosure of Invention Aiming at the problems of the traditional community service robot system in terms of understanding complex instructions, decision logic flexibility, task execution reliability and autonomous learning capacity, the invention provides an intelligent community service robot decision system based on a large model and a behavior tree. The technical scheme of the invention comprises the following steps: (1) The instruction input module is used for receiving and preprocessing a service instruction input by a user in a voice or text mode; (2) The intention analysis and task arrangement module is connected with the instruction input module and is used for judging the instruction type, calling a large language model and combining a knowledge base and an example base for control class instructions, and analyzing and arranging user instructions into a structured sub-action sequence; (3) The behavior tree construction module is connected with the intention analysis and task arrangement module and is used for automatically constructing the sub-action sequence into a behavior tree consisting of a control node, an execution node and a decoration node; (4) The behavior execution module is connected with the behavior tree construction module, and is used for controlling the robot entity to complete corresponding actions based on the scheduling and execution of the behavior tree by the robot operating system; (5) The autonomous learning module is connected with the intention analysis and task arrangement module and the behavior execution module and is used for collecting task execution samples and automatically expanding the effective samples to an example library through a double-agent evaluation mechanism; (6) The knowledge base module is connected with the intention analysis and task arrangement module and used for storing community scene environment information and robot hardware capability information. Compared with the prior art, the invention has the beneficial effects that: 1. by fusing the large language model with the behavior tree, the system can accurately understand complex and multi-step instructions of natural language expression and generate structured reliable execution logic. 2. The task flow is subjected to visual and structur