LP-MPPI: Low-Pass Filtering for Efficient Model Predictive Path Integral Control
Piotr Kicki
AI summary
Problem
Standard MPPI generates temporally uncorrelated white-noise perturbations that overrepresent high frequencies, causing inefficient search, control chattering, and excessive actuator wear.
Approach
LP-MPPI applies a low-pass Butterworth filter directly to sampled control perturbations before simulation, constraining exploration to task-relevant low frequencies while preserving rapid initial control updates.
Key results
- 24% average performance gain over SOTA MPPI variants in Gymnasium environments
- Significant reduction in high-frequency control chattering and actuator wear
- Over 32% performance improvement when integrated with Dial-MPC for quadruped locomotion
- Superior real-world racing performance on F1TENTH platforms compared to all baselines
Why it matters
Provides a computationally lightweight, interpretable enhancement to sampling-based MPC that improves control smoothness and efficiency for real-world robotic systems.
Abstract
Model Predictive Path Integral (MPPI) control is a widely used sampling-based approach for real-time control, valued for its flexibility in handling arbitrary dynamics and cost functions. However, it often suffers from high-frequency noise in the sampled control trajectories, which hinders the search for optimal controls and transfers to the applied controls, leading to actuator wear. In this work, we introduce Low-Pass Model Predictive Path Integral Control (LP-MPPI), which integrates low-pass filtering into the sampling process to eliminate detri- mental high-frequency components and enhance the algorithm’s efficiency. Unlike prior approaches, LP-MPPI provides direct and interpretable control over the frequency spectrum of sam- pled control trajectory perturbations, leading to more efficient sampling and smoother control. Through extensive evaluations in Gymnasium environments, simulated quadruped locomotion, and real-world F1TENTH autonomous racing, we demonstrate that LP-MPPI consistently outperforms state-of-the-art MPPI variants, achieving significant performance improvements while reducing control signal chattering.