# US 7,613,569 B2 · Crash prediction network with visual input for vehicle

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 7,613,569 B2
- Document ID: US7613569B2
- Status: Granted
- Kind: B2
- Application number: 11/458,261
- Filed: 2006-07-18
- Published: 2008-07-17
- Granted: 2009-11-03
- Earliest claimed: 2005-07-19
- Assignees: Toyota Motor Engineering & Manufacturing North America, Inc.; Board of Regents, The University of Texas System
- Inventors: Rini Sherony; Risto P. Miikkulainen; Kenneth O. Stanley; Nathaniel F. Kohl
- Web page: https://patentgrep.com/patent/US7613569B2

## CPC

- G06N 3/02: Neural networks
- G06V 20/58: Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- B60W 30/095: Predicting travel path or likelihood of collision
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- B60W: CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT

## Abstract

A method for facilitating the avoidance of a vehicle collision with an object includes the following steps: a) providing an environment for generating training examples, b) evolving a good driver using a visual input, c) evolving a crash predictor using a visual input, and d) outputting a warning signal.

## Claims

### Claim 1

A method for facilitating the avoidance of a vehicle collision with an object comprising the steps of:
a) providing a training environment;
b) evolving a good driver using a visual input, wherein the step of evolving a good driver includes the steps of:
e) providing an initial population of networks;
f) generating visual sensory signals;
h) inputting the visual sensory signals to-the a neural network;
h) calculating a fitness value:
i) repeating steps f-h a plurality of times for each neural network;
j) determining which neural network to discard and which neural network to reproduce;
k) forming a new population by discarding and reproducing;
l) repeating steps i-k a plurality of times;
c) evolving a crash predictor using a visual input;
d) outputting a warning signal.

### Claim 2 (depends on 1)

The method of claim 1 wherein the fitness is calculated in proportion to a distance traveled and a damage incurred over time spent off a road.

### Claim 3 (depends on 2)

The method of claim 2 wherein the sensory signals are visual pixels.

### Claim 4 (depends on 3)

The method of claim 3 including the step of scaling the visual sensory signals following the step of generating the visual sensory signals.

### Claim 5 (depends on 3)

The method of claim 3 wherein the step of scaling the visual sensory signals includes assigning each pixel a numerical value.

### Claim 6 (depends on 5)

The method of claim 5 wherein the numerical value is from zero to one.

### Claim 7 (depends on 3)

The method of claim 3 wherein the visual pixels are converted from simulated driving data.

### Claim 8 (depends on 3)

The method of claim 3 wherein the visual pixels are generated from a CMOS or CCD sensor.

### Claim 9 (depends on 1)

The method of claim 1 wherein the visual sensory signal is inputted online.

### Claim 10 (depends on 1)

The method of claim 1 wherein the visual sensory signal is inputted offline.

### Claim 11

A method for facilitating the avoidance of a vehicle collision with an object comprising the steps of:
a) providing a training environment:
b) evolving a good driver using a visual input:
c) evolving a crash predictor using a visual input wherein the step of evolving a crash predictor includes the steps of
m) providing an initial population of networks;
n) generating visual sensory signals;
o) inputting the visual sensory signals to the a neural network;
p) forming targeted outputs based on when a crash occurs;
q) calculating a fitness value;
r) repeating steps m-q a plurality of times for each neural network;
s) determining which neural network to discard and which neural network to reproduce;
t) forming a new population by discarding and reproducing;
u) repeating steps r-t a plurality of times; and
d) outputting a warning signal.

### Claim 12 (depends on 11)

The method of claim 11 wherein the fitness is calculated in proportion to a difference in the targeted output and the actual output of the network.

### Claim 13 (depends on 11)

The method of claim 11 wherein the sensory signals are visual pixels.

### Claim 14 (depends on 11)

The method of claim 11 including the step of scaling the visual sensory signals following the step of generating the visual sensory signals.

### Claim 15 (depends on 14)

The method of claim 14 wherein the step of scaling the visual sensory signals includes assigning each pixel a numerical value.

### Claim 16 (depends on 15)

The method of claim 15 wherein the numerical value is from zero to one.

### Claim 17 (depends on 13)

The method of claim 13 wherein the visual pixels are converted from simulated driving data.

### Claim 18 (depends on 13)

The method of claim 13 wherein the visual pixels are generated from a CMOS or CCD sensor.

### Claim 19 (depends on 11)

The method of claim 11 wherein the visual sensory signal is inputted online.

### Claim 20 (depends on 11)

The method of claim 11 wherein the visual sensory signal is inputted offline.