TR2023-106
Trajectory Generation for Online Payload Estimation of Robot Manipulators: A Supervised Learning Based Approach
-
- "Trajectory Generation for Online Payload Estimation of Robot Manipulators: A Supervised Learning Based Approach", IEEE Conference on Automation and Science Engineering, DOI: 10.1109/CASE56687.2023.10260415, August 2023.BibTeX TR2023-106 PDF
- @inproceedings{Duan2023aug,
- author = {Duan, Xiaoming and Wang, Yebin and Romeres, Diego and Koike-Akino, Toshiaki and Orlik, Philip V.},
- title = {Trajectory Generation for Online Payload Estimation of Robot Manipulators: A Supervised Learning Based Approach},
- booktitle = {IEEE Conference on Automation and Science Engineering},
- year = 2023,
- month = aug,
- doi = {10.1109/CASE56687.2023.10260415},
- isbn = {979-8-3503-2070-1},
- url = {https://www.merl.com/publications/TR2023-106}
- }
,
- "Trajectory Generation for Online Payload Estimation of Robot Manipulators: A Supervised Learning Based Approach", IEEE Conference on Automation and Science Engineering, DOI: 10.1109/CASE56687.2023.10260415, August 2023.
-
MERL Contacts:
-
Research Areas:
Abstract:
This paper studies the optimal trajectory generation problem for online payload estimation to enable flexible manipulation, where the robotic manipulators pick, transport, and place various types of workpieces. Prevailing work focuses on offline estimation and solve time-consuming optimization problem for the optimal trajectory and initial configuration. By contrast, online estimation requires a quick trajectory generation process where the initial configuration, largely determined by the workpiece and environment layout, is not a design variable. Parameterizing joint trajectories by sinusoidal functions with the amplitudes being design variables, we adopt a supervised learning based approach to fulfill real- time trajectory generation where the mapping from the initial joint positions to the optimal amplitudes is established. This approach shifts the burden of solving computationally intensive and time-consuming trajectory design problems offline and facilitates the fast online generation of identification trajectories. The effectiveness of the trajectory generation method is demonstrated through simulation.