PRED-MPPI: Disturbance-Preview and Efficient MPPI for Robust Quadrotor Tracking with Hardware Validation

This paper has been accepted to the 2026 IEEE International Conference on Robotics and Automation (ICRA)

Haodi Zhang^1*, Junwei Ge^1*, Jinya Su^1†, Yongping Pan¹, Jun Yang², Wen-Hua Chen³, Shihua Li¹

¹Southeast University, ²Loughborough University, ³The Hong Kong Polytechnic University

^*Contributed equally | ^†Corresponding author

Abstract

We propose PRED-MPPI, the first MPPI variant that seamlessly integrates real-time disturbance preview and adaptive discretization for quadrotor tracking control under significant model inaccuracies and time-varying disturbances. Unlike prior MPPI variants (e.g., L1-MPPI, DA-MPPI), which assume constant or matched disturbances, PRED-MPPI leverages a high order Generalized Extended State Observer for disturbance preview and a Variable Discretization Grid (VDG) to reduce computation and control variance. The synergy enables real-time (50Hz) quadrotor control under time-varying and mismatched disturbances. Extensive comparative simulation and real-world Crazyflie experiments demonstrate substantial performance gains. In AirSim simulation, PRED-MPPI reduces computation time by over 30%, and mean RMSE by 10.3%, 13.5%, and 14.6% compared to baseline MPPI, and by 2.59%, 3.62%, and 5.80% compared to DA-MPPI across three representative scenarios. In real-world Crazyflie experiment, for ground-effect-disturbed hovering, PRED-MPPI reduces mean and standard deviation (Std) of X–Y plane error by 14.2%/17.9% and 6.03%/21.6% compared to MPPI and DA-MPPI; for fan-induced wind experiments, PRED-MPPI yields improvements of 23.4%/36.8% and 13.8%/25.0% in RMSE and tracking error Std. These results establish PRED-MPPI as the first disturbance-preview MPPI achieving real-world UAV robustness and efficiency, paving the way for deployment on resource-limited robotic platforms.

Simulation Experiments

We evaluate performance of proposed and baseline algorithms on three representative trajectory primitives, zig-zag, lemniscate, and square, subjected to ramp-type disturbances that emulate time-varying external perturbations. The comparative quantitative results clearly demonstrate the superior performance of the proposed PRED-MPPI in term of computation and tracking performance, through leveraging disturbance preview technique and VDG strategy. Both simulations and experiments are conducted through Crazyflie trajectory tracking to validate the algorithm and answer the following questions:

Q1: Can the VDG strategy enhance computational efficiency of the MPPI optimization process while maintaining algorithmic performance?

Fig. 3: Average time consumption shows that MPPI leveraging VDG reduces computation time by over 30% thanks to the reduction on the number of sampled control inputs.

Q2: To what extent can PRED-MPPI outperform the MPPI and DA-MPPI under complex disturbances despite simplification in model-based trajectory sampling?

Fig. 4: Box plots of RMSE for three typical scenarios. This improvement indicates that integrating VDG in MPPI could plummet the variance of tracking error, stemming from the reduced stochastic noise due to fewer sampled inputs.

Real-world Experiments

To further answer Q1 and Q2, we further consider two scenarios for concept validation: trajectory tracking in the presence of ground effect and circular trajectory tracking under fan-induced unknown time-varying wind disturbances. In hardware experiments, ground-truth states and position observations are obtained through Virtual-Reality Peripheral Network communication with a four-camera motion capture system (ChingMu MC1300). Exogenous wind disturbances are introduced using a miniature fan.

Ground-effect-disturbed Hovering Experiments

Fig. 5: Violin plots represent the distance to (0,0) in X-Y plane of the baseline controllers W/WO VDG strategy, demonstrating the VDG strategy is able to reduce both the mean and variance of the X-Y displacement distribution.

Fig. 6: Disturbance-unaware controllers (e.g., geometry PD, MPPI, and MPPI+VDG) fail to maintain the target altitude at pz = 0.5m due to ground effect and input uncertainties, whereas the disturbance-aware controllers (e.g., DA-MPPI, DA-MPPI+VDG, PRED-MPPI) successfully achieve the target.

Geometry PD

MPPI

MPPI+VDG

DA-MPPI

DA-MPPI+VDG

PRED-MPPI

Fan-induced Wind Experiments

Fig. 7: Comparative circular trajectory tracking results in X-Y plane under fan-induced time-varying wind disturbances and input uncertainties.

Fig. 8: Comparative circular trajectory tracking results of 6 controller under fan-induced unknown time-varying wind and input uncertainties.