This research addresses the pivotal role of active anti-roll bars in mitigating
vehicle body roll during cornering, thereby enhancing overall stability,
maneuverability, and comfort. The proposed approach integrates two distinct
control methodologies—a straightforward error proportional controller and a
reinforcement learning (RL)-based controller. Each front and rear active
anti-roll bar applies a roll-reducing torque computed by the proportional
controller during cornering. However, this torque alone proves insufficient in
effectively damping roll oscillations induced by road irregularities. The
RL-based controller leverages observations encompassing inertial measurement
unit data (roll rate, pitch rate, and vertical acceleration), and wheel vertical
displacements and employs the roll as a reward signal. This controller
calculates two additional corrective torques. These torques are seamlessly
incorporated by both front and rear anti-roll bars alongside the proportional
controller, effectively minimizing cornering oscillations. The results
demonstrate the efficacy of the solution in significantly reducing vehicle roll,
even in challenging road conditions. This novel hybrid control strategy combines
the simplicity of proportional feedback with the adaptability of RL, offering a
robust anti-roll system that excels in both cornering dynamics and rough terrain
scenarios. In the test maneuver, the proportional controller showed an RMSE,
NRMSE, and MAE of 0.1626, 1.3966, and 0.9169 deg, respectively. In contrast, the
hybrid controller showed 0.0935, 1.1525, and 0.6710 deg, respectively. The
results denote a decrease in RMSE, NRMSE, and MAE of roll over null reference
between hybrid and purely proportional controller by 42.79%, 17.60%, and 27.64%,
respectively. The presented findings underscore the potential of this integrated
approach for advancing vehicle comfort, stability, and safety across diverse
driving conditions.
