Robot Collision Prediction Using Neural Scan Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot collision prediction methods require significant computational resources, especially when dealing with multiple objects, leading to increased manufacturing costs and complexity.
Innovation Solution
A robot system that uses a neural network model to predict collision possibilities by acquiring and processing scan data, including location information from sensors, and generates driving data to avoid collisions with minimal computation, utilizing techniques such as vector grouping and preprocessing to reduce computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If object recognition and tracking operations are performed on multiple objects in photographed images, then collision prediction accuracy is improved, but computational amount and device complexity increase significantly
Solution Approach 1:
The patent extracts only the essential information needed for collision prediction (object location, size, and motion trajectory) from the photographed images, rather than performing complete object recognition and tracking. This selective extraction of critical data reduces computational complexity while maintaining sufficient accuracy for collision prediction purposes.
Solution Approach 2:
The patent segments the collision prediction process into distinct stages: acquiring photographed images, extracting object information, predicting object motion, and determining collision risk. This segmentation allows each stage to be optimized independently, reducing overall system complexity while maintaining prediction accuracy.
2Reliability
If complete object recognition and tracking are performed for collision prediction, then prediction accuracy is improved, but manufacturing cost increases due to larger model size
Solution Approach 1:
The patent extracts only the minimal necessary object information (location, size, motion trajectory) required for collision prediction, rather than implementing complete object recognition systems. This extraction approach reduces the size of the prediction model and associated hardware requirements, thereby lowering manufacturing costs while maintaining sufficient prediction accuracy.
Solution Approach 2:
The patent applies partial action by performing only the specific object information extraction needed for collision prediction, rather than implementing full object recognition and tracking capabilities. This partial approach reduces system complexity and manufacturing cost while achieving the necessary prediction accuracy for safety purposes.
3Reliability
If multiple objects are processed through recognition and tracking operations, then collision detection capability is improved, but computational time and processing load increase
Solution Approach 1:
The patent extracts only the critical motion-related information (location, size, trajectory) from multiple objects in the environment, rather than performing complete recognition and tracking on all objects. This selective extraction significantly reduces computational processing time while maintaining sufficient capability to detect and predict collisions with relevant objects.
Solution Approach 2:
The patent segments the processing of multiple objects by focusing computational resources only on objects that are relevant to collision prediction, rather than uniformly processing all detected objects. This segmentation approach reduces overall computational load and increases processing speed while maintaining collision detection capability.
Data Source
AI summary
A robot includes: a sensor; a driver; a memory storing instructions; and a processor, wherein the processor is configured to execute the instructions to: acquire first scan data including location information of an object around the robot based on a sensing value acquired by the sensor at a first time point, input the first scan data to a neural network model learned to predict scan data to acquire second scan data predicted to be acquired by the robot at a second time point after the first time point, identify a possibility of collision with the object based on location information of the object included in the second scan data; acquire driving data based on the identified possibility of collision; and control the driver based on the acquired driving data.