Autonomy Safety Testing Environment

Energid Technologies proposes an alternative simulation concept for testing and validating convoy automation algorithms. Verifying large-vehicle automation completely, to the point of enabling fielding, is a big task. Due to the level of confidence needed--there must be no runaway large vehicles--it requires thoroughness that is not immediately practical using traditional hardware facilities alone. Traditional simulation, even Monte Carlo simulation, fails to completely solve this problem, as for complex scenarios it inevitably uses imperfect probability distributions for unknown parameters and inadequately samples those parameters when the dimensionality is high. Energid proposes a randomizing-optimizing technique to address these problems for convoy algorithm validation. With it, anomalies and corner cases that cause convoy failure are actively sought algorithmically. In the approach, simulation-based testing is cast as an optimization problem over a configurable metric on statistical outcomes. By optimizing for selectively bad output statistics, the parameters that give anomalous behavior are found. As proposed, this method is combined with high fidelity vehicle dynamics and terrain simulation, as well as GPU-based sensor simulations for cameras, lidar, and radar. The overall design is implemented for demonstration, with results supported through module and system validation.