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

**Authors:** David Russell, Rafael Papallas, Mehmet Dogar  
**arXiv:** [2408.11665](https://arxiv.org/abs/2408.11665)  
**Date:** 2024-08-21  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2408.11665v2/

## Key points

- Authors: David Russell, Rafael Papallas, and Mehmet Dogar
- Challenges of non-prehensile manipulation in high-dimensional systems
- Lengthy planning times associated with large state spaces
- Proposal of a novel approach based on the observation of critical degrees of freedom
- Reduction of degrees of freedom in trajectory optimization to streamline planning times
- Relevance in model predictive control (MPC) for constantly changing cost landscapes
- Simulation results showing improved performance under asynchronous MPC conditions
- Dynamic adjustment of focus on specific degrees of freedom for enhanced efficiency and effectiveness

## Summary

In their paper titled "Online state vector reduction during model predictive control with gradient-based trajectory optimisation," authors David Russell, Rafael Papallas, and Mehmet Dogar address the challenges of non-prehensile manipulation in high-dimensional systems. The main obstacle in this field is the lengthy planning times associated with large state spaces. While trajectory optimisation algorithms have shown effectiveness in various tasks, they struggle to scale to high-dimensional systems commonly found in non-prehensile manipulation scenarios involving clutter and deformable objects. The authors propose a novel approach based on the observation that different degrees of freedom become more or less critical to the task as the system evolves during manipulation. Leveraging this insight, they aim to reduce the number of degrees of freedom considered in trajectory optimisation problems to streamline planning times. This strategy is particularly relevant in the context of model predictive control (MPC), where the cost landscape of the optimisation problem is constantly changing. Through simulation results conducted under asynchronous MPC conditions, the authors demonstrate that their methods lead to improved overall performance by reducing policy lag while effectively optimizing trajectories. By dynamically adjusting the focus on specific degrees of freedom throughout manipulation tasks, their approach offers a promising solution to enhance efficiency and effectiveness in high-dimensional non-prehensile manipulation systems.