Robot cleaner
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot cleaners face inefficiencies in navigating obstacles, particularly when the height of the obstacle exceeds a reference height, leading to slowed or inaccurate cleaning operations, and struggle with maintaining positional accuracy at corners and edges.
Innovation Solution
The robot cleaner employs a deep learning-based convolution neural network (CNN) model to recognize obstacle types and determine appropriate motions, such as avoiding or climbing, and generates combined landmarks for improved positional correction using sensor data, allowing for efficient cleaning even when obstacle types are not initially recognized.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the robot cleaner operates based on distance to front-side obstacle, then it can avoid obstacles, but it has to temporarily stop operating which reduces cleaning speed and efficiency
Solution Approach 1:
The robot dynamically adjusts its cleaning pattern based on obstacle detection. When an obstacle is detected within the reference distance, it switches from straight-line cleaning to avoidance motion, and when no obstacle is present, it resumes rapid straight-line cleaning. This dynamic adaptation resolves the contradiction by maintaining high speed during safe operation while ensuring reliable avoidance when needed.
Solution Approach 2:
The robot uses sensor feedback to continuously monitor the distance to front-side obstacles and adjusts its motion accordingly. The controller receives real-time distance information and dynamically modifies the cleaning pattern, enabling the robot to maintain high cleaning speed when the path is clear while automatically slowing down to avoid obstacles when detected, thus resolving the speed-reliability trade-off.
2Adaptability or versatility
If the robot cleaner recognizes obstacle type using deep learning CNN model, then it can determine appropriate motion (avoiding or climbing), but it increases device complexity and processing time
Solution Approach 1:
The robot performs preliminary obstacle type recognition using the deep learning CNN model before executing the cleaning motion. By classifying the obstacle type in advance (e.g., determining whether it should avoid or climb), the robot can select the appropriate motion strategy beforehand, enabling more versatile obstacle response while managing system complexity through pre-processing rather than real-time complex decision-making during motion execution.
3Measurement precision
If the robot cleaner uses sensor data for positional correction at corners and edges, then it improves positional accuracy, but it requires complex landmark generation and matching algorithms
Solution Approach 1:
The patent introduces landmarks as an intermediary element for positional correction. Instead of directly complex sensor data processing for position estimation, the system uses pre-defined landmarks (features in the environment) as mediators. The robot detects these landmarks using sensors, matches them with stored landmark data, and corrects its position based on the matching results, thereby improving positional accuracy while managing algorithmic complexity through the landmark intermediary layer.
Data Source
Figure 1~2(d)
Figure 3~4
Figure 5~6
AI summary
An embodiment provides a robot cleaner comprising: a driving module for moving a cleaner body within a first cleaning area; a camera module for outputting a first and a second image obtained by photographing a front-side environment when the cleaner body is moved; and a control module for, when the type of an obstacle located in the front-side environment is recognized on the basis of the first and second image, controlling the diving module to allow the cleaner body to move while performing an avoiding motion or a climbing motion on the basis of the type of the obstacle.