Sampling-based methods have become a cornerstone of contemporary approaches to Model Predictive Control (MPC), as they make no restrictions on the differentiability of the dynamics or cost function and are straightforward to parallelize. However, their efficacy is highly dependent on the quality of the sampling distribution itself, which is often assumed to be simple, like a Gaussian. This restriction can result in samples which are far from optimal, leading to poor performance. Recent work has explored improving the performance of MPC by sampling in a learned latent space of controls. However, these methods ultimately perform all MPC parameter updates and warm-starting between time steps in the control space. This requires us to rely on a number of heuristics for generating samples and updating the distribution and may lead to sub-optimal performance. Instead, we propose to carry out all operations in the latent space, allowing us to take full advantage of the learned distribution. Specifically, we frame the learning problem as bi-level optimization and show how to train the controller with backpropagation-through-time. By using a normalizing flow parameterization of the distribution, we can leverage its tractable density to avoid requiring differentiability of the dynamics and cost function. Finally, we evaluate the proposed approach on simulated robotics tasks and demonstrate its ability to surpass the performance of prior methods and scale better with a reduced number of samples.
Sampling Strategy Design for Model Predictive Path Integral Control on Legged Robot Locomotion
School of Artificial Intelligence Shandong University Jinan, China;Department of Mechatronics;Robotics Xi'an Jiaotong-Liverpool University Suzhou, China;Robotics Xi'an Jiaotong-Liverpool University Suzhou, China University of Liverpool United Kingdom;Robotic Systems Xi'an Jiaotong-Liverpool University Suzhou, China
Reference-Free Sampling-Based Model Predictive Control
Inria - Département d’Informatique de l’École normale supérieure, PSL Research University(法国国家科学研究中心(Inria)- 工业信息学院,巴黎综合理工学院,PSL研究大学);Sorbonne Université, CNRS, Institut des Systèmes Intelligents et de Robotique, ISIR(索邦大学,法国国家科学研究中心,智能系统与机器人研究所,ISIR)
LP-MPPI: Low-Pass Filtering for Efficient Model Predictive Path Integral Control
Institute of Robotics and Machine Intelligence, Poznan University of Technology, Poznan, Poland(机器人与机器智能研究所,波兹南技术大学,波兰波兹南);IDEAS Research Institute, Warsaw, Poland(IDEAS研究所,波兰华沙)
Real-time Sampling-based Model Predictive Control based on Reverse Kullback-Leibler Divergence and Its Adaptive Acceleration
Principles of Informatics Research Division(信息学原理研究部);National Institute of Informatics (NII)(国家信息研究所);Informatics Program(信息学项目);The Graduate University for Advanced Studies (SOKENDAI)(高级研究大学(SOKENDAI));Division of Information Science(信息科学部)
Beyond Pure Sampling: Hybrid Optimization Mechanisms for Non-Convex Model Predictive Control
School of Aerospace Engineering, Georgia Institute of Technology, Atlanta, GA, USA(航空航天工程学院,佐治亚理工学院,美国亚特兰大,GA州);Development Division, Komatsu Ltd., Tokyo, Japan(Komatsu Ltd.开发部门,日本东京);Department of Electrical and Computer Engineering, Inha University, Incheon, Republic of Korea(电气与计算机工程系,inha大学,韩国仁川,大韩民国)
Stein-based Optimization of Sampling Distributions in Model Predictive Path Integral Control
Laboratory for Perceptive RObotics and Grounded REasoning SystemS(感知机器人与基础推理系统实验室);Robotics Department(机器人学系);University of Michigan, Ann Arbor(密歇根大学安娜堡分校)
Sampling-Based Model Predictive Control for Dexterous Manipulation on a Biomimetic Tendon-Driven Hand
Soft Robotics Lab, Department of Mechanical and Process Engineering, ETH Zurich(苏黎世联邦理工学院机械与过程工程系软机器人实验室);School of Computer Science, University of Leeds(利兹大学计算机科学学院)
Control of Legged Robots using Model Predictive Optimized Path Integral
University of Calgary, Department of Mechanical Engineering(卡尔加里大学机械工程系);Munich Institute of Robotics and Machine Intelligence (MIRMI), Technical University of Munich (TUM)(慕尼黑机器人与智能机械研究所(MIRMI),慕尼黑技术大学)