Automatic Tuning of Autonomous System Design Parameters for Performance Robustness

This paper proposes a method to optimize an autonomous system’s controllable design parameters in the presence of uncontrollable exogenous parameters, with the goal of tuning autonomy behaviors to perform well across a wide variety of operating conditions. The approach uses adaptive scenario generation to collect simulation-based performance datasets of the autonomy under test through variations of both the design and exogenous parameters. These datasets are used to perform model-based optimization that selects the design parameters that are robust and performant across the exogenous parameter space. This approach is analyzed using randomized test functions and a case study that calculates the optimal design parameters for an autonomous surface vessel path planner.