Game Vehicle Control via Action Prediction Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In online racing games, novice players face challenges in adapting due to the focus on reaction speed, leading to inaccurate control of motion vehicles in automatic modes, as traditional behavior tree algorithms lack interactivity and fail to accurately identify player intentions.
Innovation Solution
A method and apparatus for controlling a motion vehicle in a game that uses a trained action prediction model, combining offline reinforcement learning networks and tree models, to predict reward values for actions based on historical operation habits and adjust them personally to align with player intentions, thereby improving control accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional behavior tree algorithms are used for automatic mode control, then the control system is simple to implement, but the control accuracy and player intention identification are poor
Solution Approach 1:
The patent replaces traditional behavior tree algorithms (mechanical control system) with deep learning models including offline reinforcement learning networks and tree models (intelligent systems). This substitution enables the system to accurately predict player intentions and generate appropriate control actions, significantly improving control accuracy while maintaining manageable complexity through automated model training and inference.
Solution Approach 2:
The patent transforms the control approach by changing from fixed behavior tree parameters to dynamic parameters generated by trained deep learning models. The system uses historical operation data to train models that continuously adapt control parameters based on player behavior patterns, enabling accurate intention recognition and control without requiring complex manual configuration.
2Ease of operation
If automatic mode is provided for novice players, then the ease of operation is improved, but the interactivity and player intention recognition deteriorate
Solution Approach 1:
The patent implements feedback mechanisms where the system continuously monitors player operations and uses this information to refine control predictions. The offline reinforcement learning network and tree model are trained on historical operation data, creating a feedback loop that improves player intention recognition over time while maintaining automatic mode ease of operation.
Solution Approach 2:
The system performs preliminary actions by pre-training deep learning models on extensive historical operation data before actual game play. This preliminary training enables the system to quickly recognize player intentions and provide accurate automatic control without requiring complex real-time processing, thus maintaining both ease of operation and intention recognition capability.
Data Source
AI summary
A method for controlling a motion vehicle in a game is provided. The method includes: obtaining a direction adjustment instruction issued by the target player for the motion vehicle, and running data of the motion vehicle in real time; inputting the direction adjustment instruction and the running data into a trained action prediction model to predict a reward value of each action executed by the motion vehicle; selecting a target action from a plurality of actions according to a historical operation habit of the target player and the reward value of each action; and controlling the motion vehicle to execute the target action in the game scene. Various related systems and devices are also provided.