Autonomous Vehicle Safety Validation Under Sensor Miscalibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems lack effective methods to validate the safety of autonomous vehicle controllers under sensor miscalibration, which can lead to adverse events such as collisions, and conventional simulation techniques are not scalable or efficient.
Innovation Solution
The system employs simulations to determine performance metrics and safety thresholds by injecting miscalibration into sensor data, using a collision checker and miscalibration lookup table to assess the impact of sensor errors, allowing for the generation of scenarios that test the autonomous vehicle controller's ability to handle miscalibration without causing adverse events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional simulation techniques are used to validate safety under sensor miscalibration, then comprehensive safety validation can be achieved, but computational resources are excessive and scalability is poor
Solution Approach 1:
The system pre-computes and stores miscalibration correction values in lookup tables before actual safety validation. These lookup tables contain pre-calculated adjustments for various sensor miscalibration scenarios, allowing the collision checker to quickly apply corrections without performing complex real-time calculations during simulation execution.
Solution Approach 2:
The collision checker independently determines whether adverse events would occur by directly using simulation state data and applying miscalibration corrections from lookup tables, without requiring external validation systems. This self-contained approach eliminates the need for multiple layers of validation and reduces computational overhead.
2Reliability
If extensive simulations are performed to account for sensor miscalibration ranges, then safety coverage is improved, but simulation time and computational load increase
Solution Approach 1:
Miscalibration correction values for various sensor positions and angles are pre-computed and stored in lookup tables during system initialization. This allows the collision checker to quickly retrieve and apply appropriate corrections during simulation without performing time-consuming calculations for each scenario.
Solution Approach 2:
The system creates adjusted copies of simulation state data with miscalibration corrections applied, rather than running multiple separate simulations for each miscalibration scenario. This allows comprehensive safety validation across the full miscalibration range while using a single simulation execution.
3Measurement precision
If real-time collision checking with miscalibration correction is implemented, then safety validation accuracy is improved, but processing speed decreases
Solution Approach 1:
The collision checker pre-loads miscalibration correction data from lookup tables into memory before execution. During real-time operation, it simply retrieves pre-computed correction values based on sensor position and angle, avoiding complex real-time calculations while maintaining high accuracy in safety metric determination.
Solution Approach 2:
The system replaces complex real-time mathematical calculations with simple data retrieval operations from pre-computed lookup tables. This substitution of computational mechanics with data lookup operations maintains measurement precision while dramatically improving processing speed.
Data Source
AI summary
Techniques for determining a safety metric associated with a vehicle controller are discussed herein. To validate safe operation of a system, a simulation may be executed including determining a relative location of a simulated object within the simulation with respect to a location of a simulated vehicle, determining, based on the relative location of the simulated object, an adjusted location of the simulated object within the simulation, controlling, by the autonomous vehicle controller and based on the relative location of the simulated object, the simulated vehicle to follow a trajectory within the simulation, and performing a collision check between the simulated vehicle and the simulated object at the adjusted location. The safety metric associated with the autonomous vehicle controller may then be determined based at least in part an outcome of the collision check.