Using Imitation Learning to Implement Control Orchestration for Smart Chassis

Despite the advances in control allocation for over-actuated systems, the need for a comprehensive, optimized, and safe solution remains ongoing. Traditional methods, though mature, struggle with the complexities of coupled non-linear allocation and the need for extensive computational resources. Machine learning may provide significant advantages through its generalization and adaptation capabilities, especially in scenarios where linear approximations are employed to reduce computational burdens or when the effectiveness of actuators is uncertain. Recent advances in imitation learning, particularly behavioral cloning, and deep reinforcement learning have demonstrated promising results in addressing these challenges. This paper aims to determine the potential of using machine learning in control orchestration for smart chassis to go beyond allocation issues to include interaction management across systems, resource balance, and safety and performance limits. We present a set of techniques that we believe are relevant to experiment to address potential challenges like prediction and complexity for control allocation in smart chassis systems, which will be tested in the upcoming articles.