CN-122021645-A - Golf child robot behavior decision method based on language interaction

CN122021645ACN 122021645 ACN122021645 ACN 122021645ACN-122021645-A

Abstract

The invention relates to the technical field of artificial intelligence and robots, in particular to a method for deciding the behavior of a golf child robot based on language interaction, which is applied to the golf child robot with a mobile chassis, a carrying/ball bag mechanism, voice interaction and sensors, and comprises the following steps of voice acquisition and noise-resistant pretreatment; the method comprises the steps of golf semantic understanding and intention parameterization, context state estimation, behavior decision based on a behavior state machine, safety interlocking and action execution, round binding record and strategy updating, and analysis of natural language instructions into structural intention and parameters through a built-in domain semantic dictionary, synonymous mapping and context disambiguation mechanism, so that misunderstanding rate is greatly reduced, and interaction efficiency is improved.

Inventors

CHEN YI
LIN ZIXUAN
Xu Kaihan
SHI TIANYI
HE SHUYI
CHEN BIDE
YAO JIANING

Assignees

上海巡领科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260128

Claims (8)

1. A language interaction-based behavior decision method of a golf child robot is applied to the golf child robot with a mobile chassis, a carrying/ball wrapping mechanism and a voice interaction and sensor, and is characterized by comprising the following steps: S1, voice signal is collected through a near-field microphone array or a wearable microphone, wake-up words are identified and enter an instruction receiving window after noise reduction, echo cancellation, voice activity detection and wind noise suppression are carried out, voiceprint/equipment binding authorization is optionally executed, and a text instruction T, an authorization state AUTH and a confidence coefficient C_ asr are output; S2, performing semantic analysis on the text instruction T, outputting an intention type I and a structured parameter set P, and triggering a clarification strategy if the confidence level is lower than a threshold value, wherein the semantic analysis process introduces a golf semantic disambiguation and context correction mechanism; S3, estimating a context state X through interaction between a sensor and a history, wherein the X at least comprises a batting stage Phase, a scene area Zone, a user relative position and motion state, an environment risk state and a service load state; S4, based on behavior decision of a behavior state machine, constructing a behavior state set B, triggering state migration according to an intention type I, a structured parameter set P and a context state X, optionally introducing a behavior scoring function to select optimal behaviors, and outputting a behavior instruction Cmd; S5, performing safety interlocking and action execution, namely performing multi-factor interlocking judgment before executing the behavior instruction Cmd, executing the action if the interlocking condition is met, otherwise, entering a safety state and prompting reasons, wherein a dynamic safety zone generation mechanism based on the swing action is introduced; S6, round binding records and strategy updating, namely detecting the beginning/ending event of the ball hitting round, generating round ID and binding related data, and counting user preference and updating decision strategy parameters based on continuous rounds.
2. The method of claim 1, wherein the golf semantic disambiguation and context correction mechanism of step 2 comprises: A semantic dictionary in the golf field is built in, and comprises a club synonymous map, an action index map, a space fuzzy index map and a stage index map; Performing phase constraint disambiguation, spatial constraint disambiguation, and historical preference disambiguation based on the context state X; When the semantic confidence c_sem is below the threshold, a least disturbing clarification mode is entered, and the shot preparation phase allows only one clarification or silence to give up execution.
3. The method of claim 1, wherein the striking Phase in step 3 includes walking, selecting a pole, preparing, trying to swing, striking a ball, observing, going to a landing point, putting, and Phase determining based on a user speed change, dwell time, club swing event, user voice keyword, and location semantics.
4. The method of claim 1, wherein the behavior state set B in step 4 at least comprises following walking, side-to-side standby, quiet mode, delivery/pickup, detour avoidance, navigation to a target point, abnormal degradation, and the state transition rule at least comprises: When Phase is a preparation/trial swing/batting, forced entry into a side-rear standby/quiet mode and limited movement direction from entering the swing sector; when the I is a passing lever and the Zone is a green, limiting the running speed and selecting a low-noise mode; when the Zone is forbidden/high risk grade, forbidden to enter and select the latest reachable waiting point to prompt the user; when the user sends a distance adjustment instruction, the following distance parameters are updated in real time and the preference model is written.
5. The method of claim 1, wherein the multi-factor interlock decision in step 5 comprises an authorization interlock, a distance interlock, a zone interlock, a stability interlock, and an exception interlock, and wherein the dynamic safe zone generation mechanism comprises: generating a swing safety zone Z_safe based on the body posture and orientation of the user, the current club type, the swing stage, the historical swing amplitude and the rhythm model; Modeling by adopting one or a combination of a sector area model, an ellipsoid/polygon model and a time expansion safety zone; The behavior decision needs to meet the condition that the current position and the target path of the robot are not intersected with Z_safe, the user is immediately frozen to move and retreated to the nearest safety point when the gesture of the user is rapidly changed, and the Z_safe priority of the batting stage is higher than that of all service behaviors.
6. The method of claim 1, wherein the bound related data in step 6 includes voice intent and parameters, context status, execution behavior and logs, optionally including shot results and action analysis summaries, and wherein the preference model is updated based on successive rounds of data to achieve personalized optimization of service behavior.
7. The method of claim 1, wherein the intent type I in step 2 includes following/stopping/detouring/distance maintenance, delivering/picking/unpacking, next hole/going to a tee/going to a green, quiet mode/broadcasting mode/encouraging mode, recording rounds/starting training/rewinding summary, the structured parameter set P includes club number, target hole number/target point, distance and direction, stage preference, user preference, parameters may be null or partially default.
8. The method according to claim 1, wherein the abnormal interlocking triggering conditions in the step 5 comprise a clamping wheel, collision and abnormal sensor, and stopping immediately after triggering, and the behavior scoring function in the step4 integrates safety, efficiency, interference degree and energy consumption index to select the maximum scoring behavior.

Description

Golf child robot behavior decision method based on language interaction Technical Field The invention relates to the technical field of artificial intelligence and robots, in particular to a golf child robot behavior decision method based on language interaction. Background With the expansion of the application of robotics in sports scenes, golf children robots/ball strollers/follower robots have gradually appeared in driving ranges and courts, and their common functions include carrying ball bags, remote control or automatic following, navigation obstacle avoidance, simple voice broadcasting, etc. Meanwhile, the voice interaction technology and the large model dialogue are popular in the consumer electronics field, but the golf course is used as an outdoor special environment, has the characteristics of high noise, multiple users, multiple-stage tasks and the like, and requires that the robot can provide accompanying services like a 'child' and has the capability of understanding specific instructions (such as 'next hole', 'take 7 iron', 'help me station far spot', and the like) under the golf context, identifying batting stages and scene rules (tee boxes/fairways/greens/sand pits, and the like), and making proper behavior decisions (following, stopping, detouring, pole delivering, quiet modes, recording rounds, and the like) under the safety premise. In the prior art, the behavior control scheme of the golf child related device has obvious defects and shortcomings: The voice command lacks golf semantic understanding and parameterization, the existing scheme usually takes voice as a simple remote control switch, only can respond to basic commands such as advancing/stopping/following, and the like, and golf intention in natural language cannot be analyzed into structural parameters (such as hole number, club number, distance, stage preference, and the like), so that interaction efficiency is low and misunderstanding rate is high; The lack of behavior decisions of 'hitting stage/scene constraint' is that golf has definite stage division (walking, selecting pole, preparing, trial swing, hitting ball, etc.), the existing robot often ignores stage difference, and can approach, sound or move when a user is ready to hit ball, causing interference and even safety risk; The system lacks authorization and false triggering resistant mechanisms under the conditions of multiple users and multiple sound sources, wherein the noisy court environment, speaker speech, wind noise, ball car noise and the like are easy to cause false recognition of a robot, and the existing system lacks combination mechanisms of wake-up words, voiceprint authorization, near-field pickup, context confirmation and the like, so that false actions are frequent; The robot does not carry out interlocking control on club swing areas, human body approaching distance, gradient stability and the like when delivering a club, approaching a ball bag and avoiding obstacle detouring, so that the risks of collision, tripping and disturbing the swing exist; The existing equipment is only stopped at a mobile and object carrying layer, the voice instruction, the behavior decision, the service execution and the result record are not formed into a closed loop, and personalized service promotion cannot be realized by learning user preference; The abnormal working condition is not processed completely, and the existing system lacks an interpretable degradation strategy and a recovery strategy for the situations of lost following targets, GPS/positioning drift, occupied paths, sand pit/gradient influence, forbidden green and the like. Therefore, how to construct a special golf scene adaptive method for accurately understanding semantics, intelligently deciding based on scenes and stages and having high safety and personalized service capability becomes a technical problem to be solved currently. Disclosure of Invention Aiming at the defects in the prior art, the invention provides a behavior decision method of a golf child robot based on language interaction, which solves the problems of the related equipment of the golf child in the aspects of language interaction, behavior decision, safety control and service individuation in the prior art. The object of the invention is achieved by the following technical solutions, A golf child robot behavior decision method based on language interaction is applied to a golf child robot with a mobile chassis, a carrying/ball packaging mechanism and a voice interaction and sensor, and comprises the following steps: speech acquisition and noise-resistant preprocessing The voice signal processing method comprises the steps of S1, collecting voice signals through a near-field microphone array or a wearable microphone, S2, executing noise reduction, echo cancellation, voice Activity Detection (VAD) and wind noise suppression processing, improving the quality of the voice signals, S3, identifying wake-up words, entering an instruction