Combining modalities with different latencies for optimal motor control

Area of Science:

• Motor control and learning
• Computational neuroscience
• Robotics

Background:

• Motor tasks often involve redundant feedback signals from various sources.
• Selecting the most effective feedback loops is crucial for efficient motor learning and control.
• The influence of feedback signal characteristics, such as latency, on learning dynamics is not fully understood.

Purpose of the Study:

• To investigate the role of feedback signal latency in optimal motor control and learning.
• To propose and validate a computational framework where reinforcement learning selects influential feedback modules based on their latency.
• To examine how different feedback latencies impact module competitiveness and overall task performance.

Main Methods:

• Development of a computational framework utilizing reinforcement learning for motor control.
• Implementation of feedback modules with varying latencies.
• Testing the framework on two distinct scenarios: competing identical modules and visuomotor sequence learning with interacting sensory modalities.

Main Results:

• Reinforcement learning successfully identified and utilized feedback modules based on their latency without explicit gating.
• In visuomotor learning, a faster somatosensory module with shorter latency outperformed a slower visual module.
• Feedback latency influenced module independence and overall performance, with the faster module dominating.
• The slower visual module initially guided exploration and later enhanced robustness against noise and perturbations.

Conclusions:

• Feedback signal latency is a critical determinant for selecting influential feedback loops during motor learning.
• The proposed computational framework effectively learns to leverage available feedback for optimal motor control.
• Understanding feedback latency dynamics is essential for designing adaptive and efficient artificial motor systems.