Training system for automatic drive controlling strategies

A technology for automatic driving control and training systems, applied in the field of training systems, can solve problems such as high cost and danger

Active Publication Date: 2019-05-17
POLIXIR TECH LTD
View PDF5 Cites 17 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, the learning process of reinforcement learning requires a lot of interactive trial and error between the agent and the environment. In the actual autonomous driving scene, a lot of free exploration of the unmanned vehicle in the physical world is required. Obviously, this method is extremely dangerous. and expensive

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Training system for automatic drive controlling strategies

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0044] First, through devices such as traffic cameras, high-altitude cameras, or drones, take road videos of road vehicles, pedestrians, and non-motor vehicles in a variety of different road scenes;

[0045] Secondly, the dynamic factors in the road video are detected by manual labeling method or object detection algorithm, and the position sequence of each dynamic factor is constructed;

[0046] Then, the sequence of positions of the dynamic elements is played in the simulator, ie the activity trajectory of the dynamic elements is generated.

Embodiment 2

[0048] Embodiment 1 is to replay the action trajectory of the dynamic factors shot in the simulator. There are two defects in this approach. First, the road scene of the simulator must be consistent with the scene shot in the video. Second, the dynamic factors do not have the environment Responsiveness is simply history replaying. An improved scheme based on machine learning methods is described below.

[0049] First, take road videos through traffic cameras, high-altitude cameras, drones and other devices;

[0050]Secondly, the dynamic factors in the road video are detected by manual labeling or object detection algorithms;

[0051] Then, for each dynamic factor o, its surrounding information S(o,t) at each moment t is extracted (surrounding information includes 360-degree visible static factors around the factor and other dynamic factor information, etc.), location information L(o ,t), and pair the surrounding information S(o,t) with the location movement information L(o,t...

approach example 2

[0068] [Scheme Example 2] Simulator transfer correction migration.

[0069] First, execute the control action sequence (a1, a2, a3, ..., an) on the unmanned vehicle entity, and collect the perception state (s0, s1, s2, s3, ..., sn) after each action is executed.

[0070] Secondly, in the simulator, the initial state is set to s0, and the same action sequence (a1, a2, a3,..., an) is executed to collect the perceived state (s0, u1, u2, u3,..., un).

[0071] Then, the constructor g is used to correct the deviation of the simulator. The function g inputs the current state s and the action a=π(s) given by the control strategy π, outputs the corrected action a' that replaces the action a, and actually executes the action a' in the environment, that is, a'=g(s,a ).

[0072] Thirdly, using evolutionary algorithm or reinforcement learning method to train g, the goal is to make the unmanned vehicle entity data and the data generated by the simulator as similar as possible, that is, to...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a training system for automatic drive controlling strategies. The system is characterized by comprising the following three modules: machine learning-based simulator construction, confrontation learning-based drive control strategy searching and drive controlling strategy model migration. By means of the system, the problem that a safety and compliance control strategy cannot be obtained in the previous automatic drive field is solved.

Description

technical field [0001] The invention relates to a training system for automatic driving control strategies, which can be used for the control of unmanned equipment such as unmanned vehicles, robots, and drones, and belongs to the technical field of automatic driving. Background technique [0002] The goal of autonomous driving is to achieve a safe, compliant, and convenient personal automatic transportation system from assisting the driver to eventually replacing the driver. In existing automatic driving systems, driving control strategies are mostly schemes based on artificial rules or schemes based on real-time planning. Existing solutions do not have intelligent features, and have serious defects in realizing safe driving, and cannot design an automatic driving control strategy that can cover all scenarios, especially extreme scenarios. [0003] Recently, some autonomous driving solutions have introduced machine learning. By collecting driver driving data, the model is t...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G05B19/042G05D1/02G06K9/62G06N3/04
CPCG05D1/02G05B19/042G06N3/08G06N3/126G06N3/006G09B19/167G06F18/00G06N3/04G06N5/025G09B9/05
Inventor 秦熔均
Owner POLIXIR TECH LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap