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

arXiv: https://arxiv.org/abs/2409.16957 · HTML: https://mlsanity.com/paper/2409.16957/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Robot manipulation imitation learning, one of 13 directions in Offline RL and imitation learning from data.
- Builds on 3 of its 22 references that are in the map.
- First posted on arXiv 25 Sep 2024.

## Placed in

- Offline RL and imitation learning from data / [Robot manipulation imitation learning](https://mlsanity.com/offlinerl/robot-manipulation-imitation-learning/)

## Builds on (most cited first, 3 of 3)

- [Real-time Motion Generation and Data Augmentation for Grasping Moving Objects with Dynamic Speed and Position Changes](https://mlsanity.com/paper/2309.12547/) (2023)
- [dPMP-Deep Probabilistic Motion Planning: A use case in Strawberry Picking Robot](https://mlsanity.com/paper/2208.09074/) (2022)
- [Towards Autonomous Selective Harvesting: A Review of Robot Perception, Robot Design, Motion Planning and Control](https://mlsanity.com/paper/2304.09617/) (2023)

## Abstract (arXiv)

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%.