Towards Better Test Coverage: Merging Unit Tests for Autonomous Systems$$^{\dagger }$$

We present a framework for merging unit tests autonomous systems. Typically, it is intractable to test an system every scenario in its operating environment. The question of whether possible design single multiple requirements the motivates this work. First, we formally define three attributes test: specification that characterizes behaviors observed execution, environment, and policy. Using merge operator from contract-based theory, provide formalism construct merged two specifications. Temporal constraints on guarantee non-trivial satisfaction both specifications necessary successful execution. assume environment remains same across test. Given synthesize policy filter using receding horizon approach, use guide search procedure (e.g. Monte-Carlo Tree Search) find guaranteed satisfy specification. This finds maximizes pre-defined robustness metric while guarantees satisfying prove our algorithm sound. Furthermore, approach synthesizing ensures scalable. Finally, show impactful designing efficient campaigns achieve similar levels coverage fewer executions. illustrate self-driving examples discrete-state setting.