Mobile Robot Movement Intent Prediction via Maneuver Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Mobile robots face challenges in consistently predicting the movement intent of surrounding objects, leading to potentially unsafe situations due to limited information, such as misunderstanding traffic signals and causing accidents.
Innovation Solution
The implementation of sensors and algorithms that collect and analyze data to predict pedestrian and vehicle movement intent, including computer vision for pedestrian tracking and motion analysis for vehicle behavior, allowing mobile robots to adjust their configuration to avoid dangerous situations through V2V and V2I communication.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If mobile robots use limited sensor information to predict movement intent, then the device complexity is reduced, but the prediction reliability deteriorates leading to unsafe situations
Solution Approach 1:
The system performs preliminary classification of object movements into maneuver types (e.g., curve negotiation, zigzag, tailgating, lane changing) before making prediction decisions. This preliminary categorization allows the robot to select appropriate prediction models based on the detected maneuver type, improving prediction reliability without requiring all possible sensor information to be processed equally.
Solution Approach 2:
The system changes prediction parameters based on the classified maneuver type. Different maneuver types have different prediction models with varying parameters (e.g., curve negotiation uses lateral acceleration thresholds, tailgating uses time-to-collision calculations). This parameter adaptation allows reliable predictions using limited sensor information by focusing computational resources on relevant parameters for each specific maneuver type.
2Measurement precision
If mobile robots classify and predict multiple maneuver types, then the prediction accuracy improves, but the device complexity increases due to multiple models
Solution Approach 1:
The prediction system is segmented into multiple specialized models, each dedicated to a specific maneuver type (curve negotiation, zigzag movement, tailgating, lane changing). The classification module divides incoming movement data into these categories, and only the relevant model for each category is activated. This segmentation improves prediction accuracy for each maneuver type while managing complexity by avoiding the need to run all models simultaneously.
Solution Approach 2:
The system dynamically selects and switches between different prediction models based on the real-time classification of maneuver types. Rather than using a single static model or running all models continuously, the system adapts its prediction approach dynamically by activating only the model appropriate for the current maneuver type, thereby improving accuracy while controlling computational complexity.
3Reliability
If mobile robots adjust configuration based on predicted movement, then safety improves, but the response time is reduced due to additional processing
Solution Approach 1:
The system performs preliminary classification of maneuver types and pre-loads or pre-configures the appropriate prediction models based on this classification. By preparing the relevant prediction model in advance before actual prediction is needed, the system minimizes processing delays during critical response moments while maintaining high safety standards through accurate maneuver-specific predictions.
Data Source
AI summary
The present disclosure extends to methods, systems, and computer program products for predicting the movement intent of objects. In one aspect, a mobile robot predicts the movement intent of pedestrians from past pedestrian trajectory data and landmark proximity. In another aspect, a host mobile robot predicts the movement intent of other robots/vehicles using motion analysis models for different driving behaviors, including curve negotiation, zigzagging, rapid acceleration/deceleration, and tailgating. In a further aspect, a mobile robot can self-predict movement intent and share movement intent information with surrounding robots/vehicles (e.g., through vehicle-to-vehicle (V2V) communication). The mobile robot can self-predict future movement by comparing the operating values calculated from the monitored components to the operating limits of the mobile robot (e.g., an adhesion limit between the tires and ground). Exceeding operating limits can be an indication of skidding, oversteering, understeering, or fishtailing.