A reinforcement learning approach to gait training improves retention

Area of Science:

• Motor learning
• Biomechanics
• Neurorehabilitation

Background:

• Supervised learning with error feedback is common in gait training.
• While effective for initial adaptation, supervised learning leads to rapid forgetting of gait improvements.
• Novel approaches are needed to enhance the retention and transfer of learned gait patterns.

Purpose of the Study:

• To test if a reinforcement learning (RL) approach improves retention and transfer of a new gait pattern compared to supervised learning.
• To investigate the efficacy of self-guided exploration versus explicit instruction in motor skill acquisition.
• To evaluate the long-term retention and over-ground transfer of gait adaptations learned via RL.

Main Methods:

• Healthy subjects were randomly assigned to either a supervised group (explicit instruction, error feedback) or a reinforcement group (implicit feedback on proximity to target gait).
• Participants practiced a new gait pattern on a treadmill for 10 minutes.
• Retention was assessed immediately and after 24 hours, with over-ground transfer tests conducted.

Main Results:

• A pilot study confirmed that subjects could learn a new gait pattern using reinforcement learning.
• In a larger experiment, the reinforcement group demonstrated significantly better overnight retention (32% error increase) than the supervised group (120% error increase).
• No significant differences in over-ground transfer were observed between the two groups.

Conclusions:

• Reinforcement learning, which encourages self-guided exploration, enhances the retention of newly learned gait patterns.
• Implicit feedback paradigms may be more effective for long-term motor memory consolidation in gait training.
• Future research should explore optimizing RL parameters for gait retraining to improve both retention and transfer.