Autonomous Vehicle Behavior Prediction Using Learned Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving systems face challenges in accurately predicting the future behavior of objects in their environment, particularly when vehicle-to-vehicle communication is lacking, which affects safety and efficiency in navigating through complex scenarios like roundabouts and changing lanes.
Innovation Solution
An automated driving system that uses a perception system and computing device to detect current behavior and traffic density information, predicts future behavior by comparing it to map information and training data, and selects transitions between vehicle states to execute planned actions such as entering a roundabout or changing lanes based on a learned classifier.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If vehicle-to-vehicle communication is used to account for current behavior of objects, then the autonomous vehicle can obtain real-time behavior information, but this solution is impractical when many vehicles lack this technology
Solution Approach 1:
The patent uses perception systems (cameras, sensors) as intermediaries to observe and infer the behavior of other vehicles and objects in the environment. Instead of relying on direct communication with other vehicles, the autonomous vehicle captures images and sensor data, then uses behavior prediction models to determine current and future behavior of detected objects, making the system compatible with vehicles that lack communication technology while maintaining reliable behavior information
Solution Approach 2:
The patent replaces the mechanical/communication-based vehicle-to-vehicle communication system with a perception-based observation system. Instead of exchanging digital messages through communication protocols, the system uses optical sensors, cameras, and image processing to detect and analyze the physical state and behavior of surrounding vehicles, substituting a communication-dependent approach with a perception-based approach that works with all vehicles
2Reliability
If time to contact measurement is used to identify windows of opportunity, then the autonomous vehicle can determine safe maneuver timing, but this approach does not account for interactions between various objects in the environment
Solution Approach 1:
The patent performs preliminary behavior prediction for multiple objects in the environment before executing maneuvers. The system uses trained behavior prediction models to forecast the future behavior of detected vehicles, pedestrians, and other objects, then uses these predictions to identify safe windows of opportunity for maneuver execution. This preliminary analysis of multiple interacting objects enables safer decision-making compared to simple time-to-contact measurements
Solution Approach 2:
The patent implements a dynamic behavior prediction system that continuously updates its understanding of object behavior based on current observations and predicted future states. The system adapts to changing environmental conditions and object interactions in real-time, adjusting maneuver timing and safety assessments dynamically rather than relying on static time-to-contact calculations. This dynamic approach accounts for the complex interactions between multiple moving objects in the environment
3Measurement precision
If behavior prediction based on map information and training data is implemented, then the autonomous vehicle can accurately predict future behavior of objects, but this requires processing and comparing multiple data sources
Solution Approach 1:
The patent uses pre-trained behavior prediction models that have been trained offline on extensive training data and map information. During runtime, the system loads these pre-trained models and uses them to predict future behavior of detected objects by comparing current observations with the pre-learned patterns from training data and map information. This preliminary training approach enables accurate predictions while keeping runtime processing complexity manageable, as the heavy computational work of learning behavior patterns is performed beforehand
Data Source
AI summary
This application describes an automated driving system and methods. The automated driving system includes a perception system disposed on an autonomous vehicle. The automated driving system can detect, using the perception system, information for an environment proximate to the autonomous vehicle. The information for the environment includes current behavior of an object of interest and traffic density information. The automated driving system can also determine a classifier for the environment based on a prediction of future behavior of the object of interest. Based on the classifier, the automated driving system can identify a transition between vehicle states, the vehicle states being associated with a planned action for the autonomous vehicle. The automated driving system can also send a command to one or more vehicle systems to control the autonomous vehicle to execute the planned action according to the transition.