Autonomous Vehicle Controller Parameter Tuning via Bayesian Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional autonomous vehicle control systems face challenges in efficiently tuning controller parameters, as they often rely on human experience or brute-force search methods, which may not guarantee optimal performance and incur substantial computational costs, neglecting second-order dynamics of control subsystems like brakes, throttle, and steering.
Innovation Solution
A computer-implemented method using a Bayesian Global Optimizer with surrogate models like Gaussian Process Regress and Tree-structured Parzen Estimator to automatically tune controller parameters by simulating various driving scenarios, optimizing parameters such as time constants and adaptive gains for model-reference adaptive controllers, and lateral dynamics parameters, until predetermined performance thresholds are met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If human observers tune controller parameters based on experience, then the tuning process is simple and fast, but the optimal performance cannot be guaranteed
Solution Approach 1:
The system performs self-tuning by automatically evaluating controller parameters through simulation and using optimization algorithms to select optimal parameter sets without human intervention, thereby achieving both efficiency and optimality
Solution Approach 2:
The system implements feedback loops where controller performance is evaluated through simulation, results are fed back to the optimization algorithm, and parameters are iteratively adjusted to improve performance until optimal values are found
2Reliability
If brute-force search using grid-search algorithms is used to tune parameters, then optimal parameters can be found, but substantial computational cost is incurred
Solution Approach 1:
The system performs preliminary actions by using surrogate models to predict controller performance before running full simulations, allowing the optimization algorithm to identify promising parameter regions first and reduce the number of expensive simulations needed
Solution Approach 2:
The system introduces surrogate models as intermediaries between the parameter space and simulation results, enabling efficient exploration of the parameter space without requiring exhaustive brute-force searches
Data Source
AI summary
Systems and methods are disclosed for optimizing values of a set of tunable parameters of an autonomous driving vehicle (ADV). The controllers can be a linear quadratic regular, a “bicycle model,” a model-reference adaptive controller (MRAC) that reduces actuation latency in control subsystems such as steering, braking, and throttle, or other controller (“controllers”). An optimizer selects a set tunable parameters for the controllers. A task distribution system pairs each set of parameters with each of a plurality of simulated driving scenarios, and dispatches a task to the simulator to perform the simulation with the set of parameters. Each simulation is scored. A weighted score is generated from the simulation. The optimizer uses the weighted score as a target objective for a next iteration of the optimizer, for a fixed number of iterations. A physical real-world ADV is navigated using the optimized set of parameters for the controllers in the ADV.