Map Prior Layer for Autonomous Vehicle Behavior Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately predicting the behavior of other vehicles on the road, which can lead to safety issues due to unpredictable maneuvers that deviate from conventional rules, as existing kinematic prediction models are inadequate for systemic maneuvers.
Innovation Solution
The integration of a probability layer into the vehicle's map database, known as 'map priors,' which uses historical data to predict the behavior of other vehicles by analyzing statistical patterns and relationships between lanes, allowing the vehicle to anticipate and respond to potential illegal or aggressive maneuvers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If kinematic prediction models are used to predict other vehicles' behavior, then the system can observe and predict future positions, but the prediction accuracy deteriorates for systemic maneuvers that deviate from conventional rules
Solution Approach 1:
The system performs preliminary actions by collecting and storing historical vehicle behavior data before predictions are needed. Map priors are pre-computed from aggregated historical data at specific geographic locations, enabling the system to quickly apply learned behavioral patterns during real-time operation without needing to analyze raw data on-the-fly
Solution Approach 2:
The patent introduces map priors as an intermediary layer between the vehicle's perception system and prediction model. This intermediary contains pre-analyzed statistical patterns of vehicle behaviors at specific locations, acting as a mediator that translates raw historical data into actionable prediction probabilities for the kinematic model
2Adaptability or versatility
If the map database is updated with probability layers from historical data, then the system gains ability to predict systemic behaviors, but the device complexity increases
Solution Approach 1:
The map database is segmented into multiple layers: the base map layer containing static geographic information, and probability layers containing behavioral patterns. Each probability layer is further segmented by geographic location and vehicle maneuver type, allowing the system to access only relevant behavioral data for specific situations rather than processing entire datasets
Solution Approach 2:
The patent adds a temporal and probabilistic dimension to the traditional spatial map database. Instead of only storing static geographic coordinates, the system incorporates time-based historical behavior patterns and probability distributions, transforming the map from a purely spatial structure to a spatio-temporal probabilistic structure
3Reliability
If the autonomous vehicle adheres strictly to map rules for maneuvers, then the vehicle operates safely within defined parameters, but it cannot anticipate or respond to illegal maneuvers by other vehicles
Solution Approach 1:
The system applies preliminary anti-action by using map priors to predict potential illegal maneuvers before they occur. By analyzing historical behavioral patterns at specific locations, the system proactively identifies high-probability violation scenarios and prepares appropriate defensive responses, countering other vehicles' unpredictable behavior before it becomes a hazard
Solution Approach 2:
The patent introduces dynamics by making the vehicle's behavior adaptive rather than static. While the vehicle itself adheres to map rules, it dynamically adjusts its operational parameters based on predicted behaviors of other vehicles. The system continuously updates probability assessments and modifies its maneuver selection and timing based on real-time conditions and learned patterns
Data Source
AI summary
Systems, methods, and devices are disclosed for mapping historical information about behaviors of objects (vehicles, bicycles, pedestrians, etc.) at a location. Based on the mapped historical information, a prediction is determined about a behavior of an object proximate to an autonomous vehicle at the location, where the prediction is based on a statistical analysis of the historical information that is applied to the object. One or more behaviors of the AV are affected based on the prediction.