Autonomous Driving Scenario Retrieval Using State-Based Sensor Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicle driving systems are often overtrained on mundane data and undertrained on less common scenarios, leading to poor performance in off-nominal situations, and existing methods struggle to efficiently identify and simulate driving scenarios of interest from vast amounts of sensor data.
Innovation Solution
The system reduces three-dimensional multi-sensor data to two-dimensional time series representations, classifies them into sequences of states, and stores these as scenarios, allowing for efficient querying and simulation of autonomous driving systems to handle anomaly conditions by providing relevant training and verification data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If autonomous vehicle AI systems are trained using collected data from sensor equipped vehicles, then the training data volume increases, but the time and computational resources required to process and train on this data increases
Solution Approach 1:
The patent extracts only the essential features from raw sensor data by reducing three-dimensional multi-sensor data to two-dimensional time series representations. This extraction process isolates the critical information needed for training while discarding redundant data, thereby maintaining training data volume but reducing processing time and computational resources required.
Solution Approach 2:
The patent segments the continuous sensor data into discrete time series representations and further divides them into manageable training scenarios. By classifying time series into sequences of states and organizing them as discrete scenarios, the system enables efficient processing and selective training on relevant portions of data, reducing overall training time while maintaining data volume.
2Adaptability or versatility
If autonomous vehicle AI systems are trained on all collected driving data, then coverage of driving scenarios improves, but the system becomes overtrained on mundane scenarios and undertrained on rare off-nominal situations
Solution Approach 1:
The patent applies local quality by classifying different portions of training data with different priorities. Rare and off-nominal scenarios are identified and separated from mundane scenarios through state classification. This allows the system to provide enhanced focus and repeated exposure to critical rare scenarios while maintaining broad coverage of all scenario types, thereby improving reliability in off-nominal situations without sacrificing overall adaptability.
3Measurement precision
If three dimensional multi-sensor data is stored and processed in full detail, then measurement precision is maintained, but data processing complexity and storage requirements increase
Solution Approach 1:
The patent transforms three-dimensional multi-sensor data into two-dimensional time series representations by projecting spatial information onto a simplified coordinate system while maintaining temporal sequences. This dimensionality reduction preserves the essential measurement precision needed for training by maintaining the temporal evolution of spatial relationships, while significantly reducing data processing complexity and storage requirements.
Data Source
AI summary
Systems and methods are provided for simulating operation of an autonomous vehicle control system. Three dimensional multi-sensor data associated with a plurality of real-world drives in a sensor equipped vehicle is accessed. For a particular drive, the three dimensional multi-sensor data is reduced to a time series of two dimensional representations. The time series of two dimensional representations is classified into a sequence of states, where the sequence of states associated with the particular drive and the three dimensional multi-sensor data are stored in a computer-readable medium as a scenario. A query is received that identifies a state criteria, and the scenario is accessed based on the sequence of states matching the state criteria of the query. The three dimensional multi-sensor data of the scenario is provided to an autonomous driving system to simulate behavior of the autonomous driving system when faced with the scenario.