Autonomous Vehicle Routing Using Sparse-Location Event Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face suboptimal route planning due to sparse data at certain geographic locations, leading to increased likelihoods of undesirable events such as sudden stops or human intervention, as existing systems struggle to accurately predict operation-influencing events without sufficient training data.
Innovation Solution
A computer-implemented model is developed to infer the likelihood of operation-influencing events at geographic locations with sparse data by utilizing training data from nearby locations, employing mixed-effects models and kriging to estimate event probabilities, allowing for optimal route identification between origin and destination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If route planning relies on existing training data from geographic locations, then route optimization can be achieved for well-traveled locations, but route planning becomes suboptimal for locations with sparse or no data
Solution Approach 1:
The patent introduces an intermediary inference mechanism that bridges the gap between locations with sufficient training data and locations with sparse or no data. By using trained models to infer operation-influencing event likelihoods at data-sparse locations based on patterns learned from data-rich locations, the system enables reliable route planning across the entire geographic region without requiring extensive local data at every location.
Solution Approach 2:
The system performs preliminary training of models using aggregated data from multiple locations before deploying them for route planning. This preliminary action creates pre-trained inference models that can generalize to new locations with sparse data, allowing the system to handle data sparsity proactively rather than reactively when planning routes to under-served areas.
2Reliability
If the autonomous vehicle avoids intersections with high likelihood of operation-influencing events, then safety and comfort are improved, but route efficiency may be reduced due to longer alternative paths
Solution Approach 1:
The system dynamically changes the routing parameters by incorporating inferred likelihoods of operation-influencing events as weighted costs in the route planning algorithm. Rather than using fixed parameters like distance or time alone, the system adjusts route selection based on the probabilistic assessment of undesirable events at each intersection, allowing flexible optimization between safety and efficiency based on current conditions.
3Measurement precision
If the system collects and processes data from all geographic locations, then prediction accuracy improves, but data collection time and computational resources increase
Solution Approach 1:
The patent merges data from multiple geographic locations into a unified training dataset, combining information about operation-influencing events across different areas. This consolidation allows the system to learn generalizable patterns from aggregated data rather than requiring extensive local data at each individual location, improving prediction accuracy while reducing the total data collection burden.
Solution Approach 2:
The system creates copied or replicated models trained on aggregated data that can be deployed to infer event likelihoods at new locations without requiring direct data collection at those specific locations. This copying approach allows the system to extend prediction capabilities to underserved areas efficiently.
Data Source
AI summary
Described herein are technologies relating to computing a likelihood of an operation-influencing event with respect to an autonomous vehicle at a geographic location. The likelihood of the operation-influencing event is computed based upon a prediction of a value that indicates whether, through a causal process, the operation-influencing event is expected to occur. The causal process is identified by means of a model, which relates spatiotemporal factors and the operation-influencing events.