Research Highlight

Beyond the Bell Curve Towards Safer, Faster Control for Autonomous Systems Under Real-World Uncertainty

Real-world disturbances do not always follow Gaussian distributions. The creates a challenging trade-off for safety-critical autonomous control: simple risk models can be misleading, while more flexible alternatives can be too computationally demanding for real-time use. Professor Hyo-Sang Shin's team at KAIST developed Spectral-IRA, combining non-Gaussian uncertainty propagation, data-driven calibration, and adaptive risk allocation to turn nonlinear probabilistic safety constraints into a tractable convex control problem. Theoretical analysis and simulations show that Spectral-IRA can keep risk within prescribed limits while remaining fast enough for real-time control. ...read more