# US 11,442,455 B2 · Method and apparatus for generating local motion based on machine learning

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

- Number: US 11,442,455 B2
- Document ID: US11442455B2
- Status: Granted
- Kind: B2
- Application number: 16/672,987
- Filed: 2019-11-04
- Published: 2020-06-25
- Granted: 2022-09-13
- Earliest claimed: 2018-12-24
- Assignees: Samsung Electronics Co., Ltd.
- Applicants: Samsung Electronics Co., Ltd.
- Inventors: Joon-Kee Cho; Kyung-Rock Kim; Taesin Ha
- Web page: https://patentgrep.com/patent/US11442455B2

## CPC

- G05D 1/0214: {in accordance with safety or protection criteria, e.g. avoiding hazardous areas (monitoring the location of vehicles within a certain area, e.g. forbidden or allowed areas, in traffic control systems for road vehicles G08G1/13)}
- G05B 13/027: {using neural networks only}
- G05D 1/0221: {involving a learning process}
- G05D 1/0274: {using mapping information stored in a memory device (navigation using map-matching G01C21/30)}
- G05D: SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05B: CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS (systems for controlling or regulating non-electric variables G05D; systems for regulating electric or magnetic variables G05F; control devices or systems insofar as characterised by mechanical features only G05G)

## Abstract

Apparatuses and methods for generating a local motion of a mobile object based on a machine learning are disclosed. The methods may include determining, by processing circuitry, a local window corresponding to a current position of the mobile object in a global area; generating, by processing circuitry, a local cost map indicating a probability of a collision of the mobile object with an obstacle in the local window; generating, by processing circuitry, a local goal map indicating a local path in the local window between the current position and a local goal in the local window; and determining, by processing circuitry, a target velocity of the mobile object based on output data of a neural network, the output data based on an input of input data to the neural network, the input data including the local cost map, the local goal map, and a current velocity of the mobile object.

## Claims

### Claim 1

A method of generating a local motion for a mobile object, the method comprising:
determining, by processing circuitry, a local window as a subset of a global area such that a center of the local window represents a current position of the mobile object within the subset of the global area;
generating, by the processing circuitry, a local cost map indicating a probability of a collision of the mobile object with an obstacle in the local window;
generating, by the processing circuitry, a local goal map indicating a local path in the local window between the current position of the mobile object and a local goal in the local window;
determining, by the processing circuitry, a target velocity of the mobile object using a neural network trained to output the target velocity in response to input thereto of the local cost map, the local goal map, and a current velocity of the mobile object, wherein the neural network utilizes the pixel values within the local cost map and pixel values within the local goal map to determine the target velocity; and
controlling the mobile object to navigate through the local path from the current position to the local goal based on the target velocity determined through the trained neutral network.

### Claim 2 (depends on 1)

The method of claim 1, further comprising:
generating a global cost map indicating a probability of a collision of the mobile object with an obstacle in the global area based on map data.

### Claim 3 (depends on 2)

The method of claim 2, wherein the generating of the local cost map is based on sensor data associated with an obstacle around the mobile object and the global cost map.

### Claim 4 (depends on 2)

The method of claim 2, further comprising:
determining, by the processing circuitry, a global path in the global area between the current position and a global goal in the global area based on the global cost map.

### Claim 5 (depends on 4)

The method of claim 4, wherein the generating of the local goal map is based on the global path and the local window.

### Claim 6 (depends on 1)

The method of claim 1, wherein the neural network includes a competitive layer that operates based on pixel values of the local cost map and pixel values of the local goal map.

### Claim 7 (depends on 1)

The method of claim 1,
wherein the neural network includes,
an input layer that receives pixel values of the local cost map and pixel values of the local goal map,
a first hidden layer that receives an output of the input layer,
a second hidden layer that receives an output of the first hidden layer and the current velocity of the mobile object, and
an output layer that generates output data of the neural network based on an output of the second hidden layer.

### Claim 8 (depends on 7)

The method of claim 7, wherein the first hidden layer is a competitive layer trained based on a competitive learning.

### Claim 9 (depends on 7)

The method of claim 7, wherein nodes of the input layer are fully connected to nodes of the first hidden layer.

### Claim 10 (depends on 9)

The method of claim 9, wherein a number of the nodes of the input layer corresponds to a sum of a number of pixels in the local cost map and a number of pixels in the local goal map.

### Claim 11 (depends on 1)

The method of claim 1, wherein the neural network includes,
a convolutional layer that extracts features from the local cost map and the local goal map,
a fully connected layer that receives an output of the convolutional layer and the current velocity of the mobile object, and
an output layer that generates output data of the neural network based on an output of the fully connected layer.

### Claim 12 (depends on 1)

The method of claim 1, wherein the neural network includes,
an input layer that receives pixel values of the local cost map and pixel values of the local goal map,
a first competitive layer that receives an output of the input layer, and
a second competitive layer that generates the output data of the neural network based on an output of the first competitive layer.

### Claim 13 (depends on 12)

The method of claim 12, wherein a value in a predetermined range is represented based on a distribution of the output data of the neural network.

### Claim 14

A method of generating a local motion for a mobile object, the method comprising:
determining, by processing circuitry, a local window corresponding to a current position of the mobile object in a global area;
inputting to a neural network, by the processing circuitry, input data including a local cost map indicating a probability of a collision with an obstacle in the local window, a local goal map indicating a local path in the local window between the current position and a local goal in the local window, and a current velocity of the mobile object; and
determining, by the processing circuitry, a target velocity of the mobile object based on output data of the neural network in response to an input of the input data.

### Claim 15

An apparatus for generating a local motion for a mobile object, the apparatus comprising:
processing circuitry; and
a memory comprising an instruction to be read by the processing circuitry,
wherein when the instruction is executed by the processing circuitry, the processing circuitry is configured to,
determine a local window corresponding to a current position of the mobile object in a global area;
generate a local cost map indicating a probability of a collision of the mobile object with an obstacle in the local window;
generate a local goal map indicating a local path in the local window between the current position and a local goal in the local window; and
determine a target velocity of the mobile object based on output data of a neural network, the output data of the neural network based on an input of input data to the neural network, the input data including the local cost map, the local goal map, and a current velocity of the mobile object.

### Claim 16 (depends on 15)

The apparatus of claim 15, wherein the processing circuitry is further configured to,
generate a global cost map indicating a probability of a collision of the mobile object with an obstacle in the global area based on map data;
generate the local cost map based on sensor data associated with an obstacle around the mobile object and the global cost map; and
determine a global path in the global area between the current position and a global goal in the global area based on the global cost map.

### Claim 17 (depends on 15)

The apparatus of claim 15, wherein the neural network includes,
an input layer that receives pixel values of the local cost map and pixel values of the local goal map,
a first hidden layer that receives an output of the input layer,
a second hidden layer that receives an output of the first hidden layer and the current velocity of the mobile object, and
an output layer that generates the output data of the neural network based on an output of the second hidden layer.

### Claim 18 (depends on 15)

The apparatus of claim 15, wherein the neural network includes,
a convolutional layer that extracts features from the local cost map and the local goal map,
a fully connected layer that receives an output of the convolutional layer and the current velocity of the mobile object, and
an output layer that generates the output data of the neural network based on an output of the fully connected layer.

### Claim 19 (depends on 15)

The apparatus of claim 15, wherein the neural network includes,
an input layer that receives pixel values of the local cost map and pixel values of the local goal map,
a first competitive layer that receives an output of the input layer, and
a second competitive layer that generates the output data of the neural network based on an output of the first competitive layer.