Robot Path Planning Using Pedestrian Behavior Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing path determination methods for autonomous mobile robots often fail to effectively avoid interference with pedestrians in unpredictable environments, leading to frequent stop states and reduced merchantability, especially in crowded areas.
Innovation Solution
A path determination method that uses learned behavior models based on image data and behavior parameters to determine optimal paths for robots, considering various pedestrian walking patterns and environments, allowing the robot to smoothly navigate while avoiding interference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a possible interference time and virtual obstacle region are used to determine robot path, then interference with traffic participants is avoided, but the robot frequently stops when pedestrians move in unpredictable loci
Solution Approach 1:
The system performs preliminary actions by predicting multiple possible pedestrian paths and preparing corresponding robot path alternatives in advance. The path determination unit calculates multiple candidate paths based on predicted pedestrian behaviors before interference actually occurs, allowing the robot to smoothly transition between paths without frequent stopping.
Solution Approach 2:
The system applies dynamics by making the robot's path determination adaptive and flexible. Instead of following a fixed virtual obstacle region, the path determination unit dynamically adjusts the robot's path based on real-time pedestrian behavior predictions and actual environmental conditions, enabling smooth navigation through crowded areas.
2Ease of operation
If a grid with the lowest existence probability of obstacle is determined as path, then the robot can navigate, but the robot stops for long time when a grid with lowest existence probability is not found in crowded environments
Solution Approach 1:
The system performs preliminary path planning by pre-calculating multiple candidate paths based on predicted pedestrian behaviors. When the current path becomes unsuitable, the robot can immediately switch to a pre-prepared alternative path without waiting to find a new grid with lowest obstacle probability, thus reducing stop time in crowded environments.
Solution Approach 2:
The system changes parameters by considering multiple pedestrian walking patterns and adjusting path determination based on different behavior scenarios. Instead of relying solely on obstacle probability in static grids, the system incorporates dynamic parameters such as pedestrian velocity, direction, and predicted behavior patterns to determine optimal paths.
3Reliability
If multiple pedestrian walking patterns are considered for path determination, then the robot can smoothly navigate through crowds, but the computational complexity increases
Solution Approach 1:
The system applies segmentation by dividing the complex path determination problem into separate processing components: a behavior recognition unit that identifies pedestrian patterns, a path prediction unit that generates candidate paths based on recognized behaviors, and a path determination unit that selects the optimal path. This modular approach manages computational complexity while maintaining reliable smooth navigation.
Data Source
AI summary
A path determination method includes acquiring plural walking paths Rw of a first pedestrian M1 in a case where walking patterns of plural second pedestrians M2 are set to first to seventh walking patterns, creating plural sets of learning data of a relationship in which a mask image of a virtual robot is associated with a moving direction command for the virtual robot, the mask image and the moving direction command being obtained in a case where the virtual robot moves along each of the plural walking paths Rw in a virtual space, creating a learned model by learning model parameters of a CNN by a gradient method by using the plural sets of learning data, and determining a moving velocity command v for a robot (2) by using the learned model.