Reinforcement Learning during Locomotion

Area of Science:

• Motor learning and control
• Human locomotion and gait
• Reinforcement learning

Background:

• Learning new motor skills often involves trial-and-error exploration, increasing movement variability.
• In gait, increased variability can compromise balance and safety, potentially limiting reinforcement learning effectiveness.
• Understanding how reinforcement learning applies to acquiring novel locomotor patterns is crucial.

Purpose of the Study:

• To investigate if humans can acquire and retain a novel locomotor pattern using reinforcement learning alone.
• To compare learning, motor variability, and motor memory formation between reinforcement learning and target error correction groups.
• To determine the impact of reinforcement learning on explicit and implicit motor memories during gait acquisition.

Main Methods:

• Participants learned a novel stepping pattern on a treadmill using binary reward feedback (reinforcement learning group).
• A comparison group learned the same pattern by correcting for target error using real-time visual feedback.
• Learning, motor variability, and explicit/implicit motor memories were assessed in two experiments.

Main Results:

• The reinforcement learning group acquired the novel walking pattern through exploration, characterized by increased motor variability.
• Reinforcement learning did not enhance implicit motor memories but led to more accurate explicit motor memories compared to the target error group.
• Participants retained a significant portion of the learned walking pattern over 24 hours.

Conclusions:

• Humans can successfully acquire new walking patterns via reinforcement learning, utilizing exploration.
• Reinforcement learning effectively enhances explicit motor memories for novel gait patterns.
• The acquired motor skills demonstrate considerable retention over a 24-hour period.