Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Online state vector reduction during model predictive control with gradient-based trajectory optimisation

David Russell, Rafael Papallas, Mehmet Dogar

TL;DR

High‑dimensional non‑prehensile manipulation suffers from long trajectory planning times; the paper presents online state‑vector reduction within an MPC framework to keep the optimization low‑dimensional while the full dynamics are evaluated in a simulator. It implements iLQR‑SVR and uses $K$‑informed metrics (sum and SVD) to selectively retain or drop DoFs, reintroducing others as needed, enabling asynchronous MPC. Across three high‑dimensional tasks in clutter and with deformable objects, the approach reduces MPC_Cost and optimization time, thereby lowering policy lag and enabling real‑time closed‑loop control without relying on task‑specific offline models.

Abstract

Non-prehensile manipulation in high-dimensional systems is challenging for a variety of reasons. One of the main reasons is the computationally long planning times that come with a large state space. Trajectory optimisation algorithms have proved their utility in a wide variety of tasks, but, like most methods struggle scaling to the high dimensional systems ubiquitous to non-prehensile manipulation in clutter as well as deformable object manipulation. We reason that, during manipulation, different degrees of freedom will become more or less important to the task over time as the system evolves. We leverage this idea to reduce the number of degrees of freedom considered in a trajectory optimisation problem, to reduce planning times. This idea is particularly relevant in the context of model predictive control (MPC) where the cost landscape of the optimisation problem is constantly evolving. We provide simulation results under asynchronous MPC and show our methods are capable of achieving better overall performance due to the decreased policy lag whilst still being able to optimise trajectories effectively.

Online state vector reduction during model predictive control with gradient-based trajectory optimisation

TL;DR

High‑dimensional non‑prehensile manipulation suffers from long trajectory planning times; the paper presents online state‑vector reduction within an MPC framework to keep the optimization low‑dimensional while the full dynamics are evaluated in a simulator. It implements iLQR‑SVR and uses ‑informed metrics (sum and SVD) to selectively retain or drop DoFs, reintroducing others as needed, enabling asynchronous MPC. Across three high‑dimensional tasks in clutter and with deformable objects, the approach reduces MPC_Cost and optimization time, thereby lowering policy lag and enabling real‑time closed‑loop control without relying on task‑specific offline models.

Abstract

Non-prehensile manipulation in high-dimensional systems is challenging for a variety of reasons. One of the main reasons is the computationally long planning times that come with a large state space. Trajectory optimisation algorithms have proved their utility in a wide variety of tasks, but, like most methods struggle scaling to the high dimensional systems ubiquitous to non-prehensile manipulation in clutter as well as deformable object manipulation. We reason that, during manipulation, different degrees of freedom will become more or less important to the task over time as the system evolves. We leverage this idea to reduce the number of degrees of freedom considered in a trajectory optimisation problem, to reduce planning times. This idea is particularly relevant in the context of model predictive control (MPC) where the cost landscape of the optimisation problem is constantly evolving. We provide simulation results under asynchronous MPC and show our methods are capable of achieving better overall performance due to the decreased policy lag whilst still being able to optimise trajectories effectively.
Paper Structure (6 sections, 13 equations, 1 figure, 1 algorithm)

This paper contains 6 sections, 13 equations, 1 figure, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Problem definition
  4. Definitions
  5. Method
  6. Optimise

Figures (1)

  • Figure 1: A sequence of snapshots showing an example MPC trajectory generated by our method. The task is to push the green cylinder to a goal region (the green transparent cylindrical region) whilst minimally disturbing some clutter objects. The full number of DoFs in this system is 55. Our method identifies the relevant DoFs of this system at different times during execution and performs trajectory optimisation using this reduced state. Objects with stronger shades of red have more DoFs in the state vector at that point during execution: If an object is dark red, all of its six DoFs are considered; if an object is white, none of its six DoFs are considered during trajectory optimisation.