Virtual Vehicle Control Using Player-Trained Driving Policies

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The existing methods for simulating virtual vehicle control in game applications require manual formulation of traveling logic, leading to long development times, high costs, and inflexible simulation effects.

Innovation Solution

A virtual vehicle control method using machine learning algorithms to train a behavior decision-making model based on player-controlled video data, allowing for automated control of virtual vehicles that mimic actual player habits, reducing the need for manual input and improving simulation realism.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If manual formulation of traveling logic is used, then the virtual vehicle control can be implemented, but the development time and costs increase significantly

Engineering Contradiction:
Improvedevelopment efficiencyVSAvoiddevelopment time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent copies actual player control behaviors from video game recordings to create training data for the machine learning model. Instead of manually creating travel logic, the system captures and reproduces real player actions, making the development process more efficient and realistic

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent replaces the manual mechanical process of formulating travel logic with an automated machine learning system. The FSM/BT manual configuration is substituted by an automated model training process that learns from video data, significantly reducing development time and effort

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If manual formulation of traveling logic is used, then the virtual vehicle control can be implemented, but the simulation effect becomes inflexible and poor

Engineering Contradiction:
Improvesimulation realismVSAvoidflexibility of travel logic
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent makes the travel logic dynamic by training a machine learning model that can adapt to different situations. Instead of fixed manual rules, the model learns patterns from video data and can flexibly respond to various game scenarios, improving both realism and adaptability

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the parameters of the control system from fixed manual rules to learned probabilistic patterns. The machine learning model outputs control actions based on learned parameters from training data, allowing the system to adapt to different situations while maintaining realistic player-like behavior

Inventive Principle:
Principle #35Parameter changes

3Reliability

If training samples are extracted from video data, then the application's internal data is protected, but the data extraction process becomes more complex

Engineering Contradiction:
Improvedata securityVSAvoiddata extraction complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts only the necessary visual information from video data without accessing or invading the application's internal data structures. By taking out only the required training samples from the video output, the system maintains data security while obtaining sufficient training material

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent uses video data as an intermediary between the game application and the training system. Instead of directly accessing internal game data, the video serves as a safe intermediary that contains all necessary visual information for training while protecting the application's internal data integrity

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3796112B1Virtual vehicle control method, model training method, control device and storage medium
Publication Date: 2022.08.24 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP3796112B1 patent drawingFigure 1~2
  • EP3796112B1 patent drawingFigure 3~4
  • EP3796112B1 patent drawingFigure 5~6

AI summary

Disclosed is a virtual vehicle control method. The method comprises: acquiring a virtual vehicle travelling picture (1101), wherein the virtual vehicle travelling picture (1101) contains a picture about a virtual vehicle travelling on a virtual racetrack; extracting a virtual racetrack image (1102) from the virtual vehicle travelling picture (1101), wherein the virtual racetrack image (1102) is an image of a virtual racetrack section where the virtual vehicle is located; inputting the virtual racetrack image (1102) into a travelling decision model, wherein the travelling decision model is generated by means of training according to a racetrack sample image and a travelling sample policy, the racetrack sample image and the travelling sample policy are extracted from a vehicle travelling sample picture, and the vehicle travelling sample picture contains a picture about a virtual vehicle, controlled by a real player, travelling on the virtual racetrack; and controlling the travelling of the virtual vehicle according to a target travelling policy output by the travelling decision model.