Related Experiment Video
Updated: Jul 13, 2026
08:16
Adaptation of a Haptic Robot in a 3T fMRI
Published on: October 4, 2011
9.7K
Velocity-aware spatial-temporal attention LSTM model for inverse dynamic model learning of manipulators
Wenhui Huang1, Yunhan Lin1, Mingxin Liu1
1Institute of Robotics and Intelligent Systems, Wuhan University of Science and Technology, Wuhan, Hubei, China.
Frontiers in Neurorobotics
|February 26, 2024
Summary
A new Velocity Aware Spatial-Temporal Attention Residual LSTM (VA-STA-ResLSTM) neural network improves manipulator inverse dynamics modeling accuracy significantly. This advanced model enhances feature extraction from motion sequences for better robotic control.
Area of Science:
- Robotics
- Artificial Intelligence
- Machine Learning
Background:
- Accurate inverse dynamics models are crucial for manipulator control.
- Existing neural network approaches face challenges with spatiotemporal variations in motion sequences.
Purpose of the Study:
- To propose a novel neural network architecture for learning more accurate manipulator inverse dynamics models.
- To address the impact of spatiotemporal variations on network learning.
Main Methods:
- Development of the Velocity Aware Spatial-Temporal Attention Residual LSTM (VA-STA-ResLSTM) network.
- Utilizing a velocity-aware spatial-temporal attention mechanism for selective feature extraction.
- Implementing a multi-layer perception attention mechanism and velocity-aware state fusion for LSTM hidden units.
Main Results:
- Achieved average accuracy improvements of 61.88% and 43.93% on two open datasets, and 71.13% on a self-built dataset compared to standard LSTM.
- Demonstrated a 10% higher modeling accuracy compared to state-of-the-art methods.
- Visualizations confirmed reliance on partial features for dynamic modeling.
Conclusions:
- The VA-STA-ResLSTM network significantly enhances the accuracy and generalization of manipulator inverse dynamics models.
- The velocity-aware attention mechanism effectively extracts relevant spatiotemporal features.
- Findings provide insights for optimizing future inverse dynamics model learning methods.
