Robot Route Planning Using Pedestrian Movement Pattern Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing route determination methods for autonomous mobile robots often result in frequent interference and halts due to unpredictable movements of pedestrians, leading to low productivity, especially in crowded environments, as they struggle to accurately avoid obstacles and determine optimal routes.
Innovation Solution
A method that generates learning data by recognizing movement routes while avoiding interference, using environmental images with color-coded displacement behaviors and action parameters to create a learned model for determining target movement routes, allowing the robot to move smoothly to a destination while avoiding multiple moving objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional route determination methods using predicted interference likelihood times and virtual obstacle regions are used, then interference avoidance capability is improved, but robot halts frequently and productivity decreases
Solution Approach 1:
The system performs preliminary actions by pre-calculating multiple possible movement patterns for surrounding objects and pre-determining multiple candidate routes for the robot in advance. This allows the robot to have ready-made alternative routes when interference is detected, reducing the need for frequent halts and improving productivity while maintaining interference avoidance capability.
Solution Approach 2:
The system dynamically adapts route determination by continuously monitoring actual movement patterns of surrounding objects and adjusting the selected route in real-time. When pedestrians deviate from predicted patterns, the system dynamically switches between pre-calculated routes, making the route determination flexible and responsive to changing conditions, thereby reducing unnecessary halts.
2Reliability
If grid squares with lowest obstacle probability are determined using A* search algorithm, then route planning capability is improved, but robot halts are lengthened in crowded environments
Solution Approach 1:
Multiple candidate routes are calculated in advance using the A* search algorithm before the robot needs to navigate. This preliminary route preparation allows the robot to quickly switch between pre-computed routes without performing time-consuming calculations during halts, thereby reducing halt time while maintaining reliable route planning capability.
Solution Approach 2:
The system calculates more routes than strictly necessary (excessive action) by generating multiple candidate routes in advance. This ensures that when the robot encounters crowded environments or unpredictable pedestrian movements, suitable alternative routes are already available, reducing the need for prolonged halts to recalculate routes.
3Measurement precision
If predicted interference likelihood times are calculated based on traffic participant velocities, then interference prediction is improved, but accuracy decreases when pedestrians move along unpredictable trajectories
Solution Approach 1:
The system dynamically adjusts interference prediction by continuously comparing actual pedestrian movements against multiple predicted movement patterns. When pedestrians deviate from predictable trajectories, the system adapts by selecting from pre-calculated routes that account for various movement scenarios, maintaining prediction accuracy while handling unpredictable behaviors.
Solution Approach 2:
The system changes the approach from relying on single velocity-based predictions to considering multiple movement patterns with different parameters. By pre-calculating routes based on various possible movement scenarios and adjusting which route to follow based on actual observed behavior, the system improves both prediction accuracy and adaptability to unpredictable pedestrian movements.
Data Source
AI summary
In an environment in which a plurality of second pedestrians moves along predetermined movement patterns, a plurality of movement routes Rw when a first pedestrian moves toward a destination point is recognized. Data, in which an environmental image indicating a visual environment in front of a virtual robot when the virtual robot moves along each of the movement routes and a moving direction command indicating a moving direction of the virtual robot are combined, is generated as learning data. In the environmental image, colors corresponding to time-series displacement behaviors of a moving object image region is applied to at least a portion of the moving object image region indicating a pedestrian (moving object) present around a robot. Model parameters of a CNN (action model) is learned using the learning data, and a moving velocity command for a robot is determined using a learned CNN.