Related Experiment Videos
Equilibrium point control of a monkey arm simulator by a fast learning tree structured artificial neural network
1Cognitive Processes Department, ATR Auditory and Visual Perception Research Laboratories, Kyoto, Japan.
Biological Cybernetics
|January 1, 1993
Summary
A neural network efficiently learned complex arm movement control, replacing a slow biomechanical model. This biologically plausible method enables faster, more adaptable robotic and prosthetic limb control.
Area of Science:
- Robotics and Biomechanics
- Computational Neuroscience
- Artificial Intelligence
Background:
- Simulating monkey arm movements requires complex biomechanical models.
- Existing models like the equilibrium point hypothesis can be computationally intensive.
- The backdriving algorithm, while effective, is slow and lacks memory.
Purpose of the Study:
- To develop a computationally simpler and more biologically plausible control method for arm movements.
- To train a neural network to replicate the functionality of the backdriving algorithm.
- To improve the efficiency and adaptability of robotic arm control systems.
Main Methods:
- A planar 17 muscle model of a monkey's arm was simulated.
- A fast learning, tree-structured neural network was trained.
- The network learned the nonlinear mapping from hand position to motor commands.
Main Results:
- The neural network learned to generate motor commands for arm movements efficiently.
- Training on 20 trajectories took only 20 minutes on a Sun-4 Sparc workstation.
- The network successfully generated accurate motor commands for both trained and untrained trajectories.
Conclusions:
- A neural network offers a computationally efficient and biologically plausible alternative to complex biomechanical models for motor control.
- This approach significantly reduces computation time compared to traditional methods.
- The trained network demonstrates adaptability and potential for real-time applications in robotics and prosthetics.