VIMPPI: Enhancing Model Predictive Path Integral Control with Variational Integration for Underactuated Systems

TL;DR

The paper addresses controlling underactuated, chaotic double pendulum systems where traditional MPPI planning is limited by short horizons. It introduces VIMPPI, which substitutes a variational integrator for the explicit Euler step in MPPI rollouts, enabling a 4–20x longer planning horizon at the same computational cost and achieving a high-frequency real-time controller at $500$–$700$ Hz. The method is augmented with linear control interpolation and a disturbance-warm-start strategy, and validated on pendubot and acrobot, where VIMPPI significantly outperforms baselines and other MPPI variants. This approach demonstrates the practical impact of numerically stable, energy-momentum-preserving integration for real-time, long-horizon planning in resource-constrained robotic systems, with potential extensions to humanoids, quadrupeds, and multi-link manipulators.

Abstract

This paper presents VIMPPI, a novel control approach for underactuated double pendulum systems developed for the AI Olympics competition. We enhance the Model Predictive Path Integral framework by incorporating variational integration techniques, enabling longer planning horizons without additional computational cost. Operating at 500-700 Hz with control interpolation and disturbance detection mechanisms, VIMPPI substantially outperforms both baseline methods and alternative MPPI implementations

Figures (2)

• Figure 1: Time series results for VIMPPI-AR-EAPO controller on the pendubot.
• Figure 2: Time series results for VIMPPI-AR-EAPO controller on the acrobot.