Towards Trajectory Conflict Prediction Using AI/ML For V&V Test Case Generation

System Verification and Validation Testing (V&V) for time-dependent systems requires the generation of test cases. Each test case is defined by a set of initial conditions and an expected outcome at the end of the specified time period. Traditional methods for generating V&V test-cases run simulations of the system to generate outcomes for each combination of initial conditions. Due to the combinatorics of even a small set of initial conditions, covering the complete combinatorics can be time and/or cost prohibitive. This paper evaluates the feasibility of using Deep Learning Neural Networks (DLNN) to generate additional test cases that were not generated by the simulations due to time limitation. A DLNN trained to on the subset of test-cases from the simulation, learns the underlying behavior of the system, and is used to generated additional test cases. A case study for using DLNN to predict test-cases for trajectory conflicts demonstrates the feasibility of this approach for time-dependent systems that exhibit bounded, deterministic behavior. The implications of these results, the limitations, and future work are discussed.