Stanford researchers slash MPPI computation time 3x with entropy-driven cooling

This paper presents an Information-Theoretic Adaptive Cooling (ITAC) framework for deterministic Model Predictive Path Integral (MPPI) control. ITAC uses the Shannon entropy of importance weights as an online feedback signal to dynamically regulate the temperature, adapting the cooling rate based on the current sampling state. The authors prove asymptotic convergence to the deterministic optimum and derive a critical entropy threshold that prevents premature weight collapse. Experiments on nonsmooth signal temporal logic motion-planning tasks demonstrate that ITAC improves sampling efficiency and achieves substantially faster convergence than state-of-the-art baselines while preserving the derivative-free nature of MPPI.