Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DualLQR: Efficient Grasping of Oscillating Apples using Task Parameterized Learning from Demonstration

Robert van de Ven, Ard Nieuwenhuizen, Eldert J. van Henten, Gert Kootstra

TL;DR

Robotic selective harvesting must safely and efficiently grasp oscillating fruits, a challenge when final approach requires close tracking but overall path length should be minimized. The authors introduce DualLQR, a dual task-parameterized framework that runs finite-horizon LQR controllers in two reference frames and fuses their outputs, avoiding expensiveLQR refitting while emphasizing the frame needing higher precision. Across simulations, DualLQR improves final-approach accuracy and reduces travel distance relative to InfLQR and a single-frame LQR, and it achieves a 99% success rate in a real apple grasping task. This approach offers a scalable method for robust, efficient autonomous fruit harvesting and can be extended to incorporate additional reference frames and obstacle considerations.

Abstract

Learning from Demonstration offers great potential for robots to learn to perform agricultural tasks, specifically selective harvesting. One of the challenges is that the target fruit can be oscillating while approaching. Grasping oscillating targets has two requirements: 1) close tracking of the target during the final approach for damage-free grasping, and 2) the complete path should be as short as possible for improved efficiency. We propose a new method called DualLQR. In this method, we use a finite horizon Linear Quadratic Regulator (LQR) on a moving target, without the need of refitting the LQR. To make this possible, we use a dual LQR set-up, with an LQR running in two separate reference frames. Through extensive simulation testing, it was found that the state-of-art method barely meets the required final accuracy without oscillations and drops below the required accuracy with an oscillating target. DualLQR, on the other hand, was found to be able to meet the required final accuracy even with high oscillations, while travelling the least distance. Further testing on a real-world apple grasping task showed that DualLQR was able to successfully grasp oscillating apples, with a success rate of 99%.

DualLQR: Efficient Grasping of Oscillating Apples using Task Parameterized Learning from Demonstration

TL;DR

Robotic selective harvesting must safely and efficiently grasp oscillating fruits, a challenge when final approach requires close tracking but overall path length should be minimized. The authors introduce DualLQR, a dual task-parameterized framework that runs finite-horizon LQR controllers in two reference frames and fuses their outputs, avoiding expensiveLQR refitting while emphasizing the frame needing higher precision. Across simulations, DualLQR improves final-approach accuracy and reduces travel distance relative to InfLQR and a single-frame LQR, and it achieves a 99% success rate in a real apple grasping task. This approach offers a scalable method for robust, efficient autonomous fruit harvesting and can be extended to incorporate additional reference frames and obstacle considerations.

Abstract

Learning from Demonstration offers great potential for robots to learn to perform agricultural tasks, specifically selective harvesting. One of the challenges is that the target fruit can be oscillating while approaching. Grasping oscillating targets has two requirements: 1) close tracking of the target during the final approach for damage-free grasping, and 2) the complete path should be as short as possible for improved efficiency. We propose a new method called DualLQR. In this method, we use a finite horizon Linear Quadratic Regulator (LQR) on a moving target, without the need of refitting the LQR. To make this possible, we use a dual LQR set-up, with an LQR running in two separate reference frames. Through extensive simulation testing, it was found that the state-of-art method barely meets the required final accuracy without oscillations and drops below the required accuracy with an oscillating target. DualLQR, on the other hand, was found to be able to meet the required final accuracy even with high oscillations, while travelling the least distance. Further testing on a real-world apple grasping task showed that DualLQR was able to successfully grasp oscillating apples, with a success rate of 99%.
Paper Structure (23 sections, 19 equations, 7 figures, 1 table)

This paper contains 23 sections, 19 equations, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background
  3. Task-Parameterization
  4. InfLQR
  5. Before execution
  6. During execution
  7. Single LQR
  8. Before execution
  9. During execution
  10. Proposed method
  11. Before execution
  12. During execution
  13. Experimental set-up
  14. Set-up of simulation experiment
  15. Set-up of apple grasping experiment
  16. ...and 8 more sections

Figures (7)

  • Figure 1: Flowcharts of the pre-process and real-time processes of DualLQR
  • Figure 2: Typical example of a demonstration. The wireframes indicate the start and end-frame. The purple line indicates the start of the final approach.
  • Figure 3: Example of final approach. The wireframes indicate the end-frame. The black line indicates the trajectory. The purple line at 5 cm from the target indicates the start of the final approach. The orange lines indicate the boundaries. The trajectory is dashed green if it lies within the boundaries. Else, it is dashed red.
  • Figure 4: Initial state of the apple grasping set-up
  • Figure 5: Results of the simulation testing of both methods. Darker lines indicate higher oscillation levels. The levels are described in section \ref{['sec:sim-exp']}. The figures on the left show the effect of an oscillation of the target's position, and the figures on the right show the effect of an oscillation of the target's orientation. At each tested setting, a vertical bar indicates the 95% confidence interval. The black dashed line indicates the required accuracy, set at 0.88.
  • ...and 2 more figures