Research Analyzer
← Back ICRA 2026

Real-Time Model Predictive Control of Nonlinear Coupled Joints Using MPPI: Application to Humanoid Ankle Joints

Gunoo Park, Jaewan Bak, Yunsoo Seo, Euncheol Im, Hoseok LEE, Jongbok Lee, Nicolas Mansard, Jongwon Lee, Yisoo Lee

PDF

AI summary

Key figure (auto-extracted from paper)
A sampling-based MPC controller combined with a lightweight neural network mapping enables real-time, constraint-aware control of highly nonlinear coupled robotic joints.
Model Predictive Path Integral Nonlinear Coupled Joints Real-Time Control Humanoid Robotics Neural Network Mapping Parallel Mechanisms

Problem

Parallel robotic mechanisms like humanoid ankle joints exhibit complex nonlinear coupling that traditional controllers struggle to handle in real time while safely enforcing joint limits.

Approach

The method uses Model Predictive Path Integral (MPPI) control to optimize actuator commands, integrating a trained neural network that maps actuator positions to joint space for accurate, constraint-aware cost evaluation.

Key results

  • First real-time MPC controller for nonlinear coupled parallel joints
  • Lightweight neural network maps actuator-to-joint kinematics with ~0.0019 ms inference latency
  • MPPI controller accurately tracks step and spline commands while strictly respecting joint limits
  • Demonstrated safe, stable control on a physical 2-DOF humanoid ankle joint platform

Why it matters

Enables safe, agile, and computationally efficient control of complex parallel robotic mechanisms, advancing dynamic humanoid leg design and locomotion.

Abstract

Modern robotic systems increasingly employ non- linear coupled joints, which present significant challenges in control. Unlike traditional serial chain configurations, where simplicity was the primary concern, parallel mechanisms such as those found in humanoid ankle joints add another layer of complexity. In this work, we propose an actuation controller for nonlinear coupled joints based on Model Predictive Path Integral (MPPI) control framework: a sampling-based model predictive control framework that incorporates nonlinearity and coupling effect simultaneously. Highly nonlinear Actuator- Joint mapping, expressed through lightweight neural network, enables intuitive controller design by exposing the actuator space control to the joint space command. Also, our method enables posing joint limit constraints, enabling safe operation on a real-robot platform. To experimentally validate our method, joint position control of a humanoid ankle joint with 2-DOF has been conducted, where accurate, real-time control and constraint-respecting behavior has been demonstrated.

Index terms

Actuation and Joint Mechanisms Optimization and Optimal Control Humanoid Robot Systems

Related papers