Final Report Summary - EMLNPRGT (Enhancing motor learning and neural plasticity in robotic gait training)
Recent work in robot-aided training has focused on developing sophisticated robotic mechanisms in order to support movement training of complex movements, such as walking. Although robot-aided gait training has been presented as a promising technique to improve motor learning and rehabilitation, to date, the learning gains obtained after training are limited. Robotic guidance is generally used in motor training to reduce performance errors while practicing. However, there is little evidence that robotic guidance enhances motor learning. In fact, research on motor learning has emphasized that errors are fundamental signals that drive motor adaptation. Thereby, robotic training strategies that amplify movement errors rather than decrease them might have a great potential to provoke better motor learning, especially in initially more skilled subjects.
The proposed research project aimed to develop control algorithms in order to enhance motor learning and neural plasticity in robot-aided gait training. The hypothesis behind the research project is that learning a locomotor task can be optimized when errors are decreased or amplified based on subjects’ initial skill level. Therefore, the first project’s objective consisted in performing an evaluation of the effect that the individual subjects’ performance might have on the effectiveness of different forms of robotic training algorithms that amplify or reduce movement errors on motor learning of a locomotor task. The goal of robotic therapy is to develop robotic devices in order to perform exercises which provoke motor plasticity. However, it is still an open question how different rehabilitation strategies contribute to brain restorative processes. Therefore, the second objective of the research project consisted in an evaluation of the brain regions involved in learning when training with different forms of guidance, by means of performing functional Magnetic Resonance Imaging (fMRI) while training with an fMRI compatible walking robotic device.
The first step towards the project completion consisted in developing new control strategies to improve motor learning of locomotor tasks using the MRI-compatible robotic device developed in the host institution (MARCOS). Five different training strategies that reduce or amplify movements errors during training were developed: i) Haptic guidance: errors were eliminated by passively moving the limbs, ii) No guidance: no robot disturbances nor robotic guidance were presented, iii) Error amplification: existing errors were amplified with repulsive forces, iv) Noise disturbance: errors were evoked intentionally with a randomly-varying force disturbance on top of the no guidance strategy, and v) Assist-as-needed: a compliant controller capable to adapt the robot assistance step by step until the value of the assistance converges into an amount slightly lower than the least amount of assistance the subject needs to achieve the desired movement. The new controllers’ performance was experimentally tested inside and outside the MRI scanner and they were proved to provide accurate pneumatic control in MR environments in order to allow assessments of brain activation.
A first motor learning study with 22 healthy subjects was run in order to investigate the effect that the subjects’ individual skill level had on the effectiveness of the designed training modes that amplify or reduce errors on motor learning of a simple locomotor task. Subjects were instructed to relax the dominant leg while it was moved by the robot, and to synchronize the non-dominant leg with the passive leg to achieve a gait-like alternating movement with same amplitude and frequency. A within-subject cross-over design was used to test the effects of training with four different training modes: i) Haptic guidance, ii) No Guidance, iii) Error amplification, and iv) Noise force disturbance. Results indicated that training strategies that reduce or do not amplify errors limit muscle activation during training and result in poor learning gains. Adding random disturbing forces during training seems to increase attention, and therefore improve motor learning. Error amplification was the most difficult strategy for the subjects, aroused the highest muscle activation, brought an after effect and also seemed to provoke the best motor learning in less skilled subjects, perhaps because subjects could better detect their errors and correct them. Skilled subjects could not improve their performance, probably because the task was too easy.
Results from this motor experiment contradicted recent studies that suggested that training with appropriately designed haptic guidance can enhance learning of some motor skills, such as time-critical tasks. Walking is a time-critical task, since it involves a certain level of rhythmicity, and therefore a positive effect of haptic guidance was expected. However, no studies to date have systematically evaluated the effect that haptic guidance have in learning to perform rhythmic tasks, compared to discrete movements (i.e. tasks that include well-defined postures at the beginning and the end of the movement, such as target-oriented reaching movements). In order to investigate the effect of haptic guidance (and no guidance) on learning time-critical continuous rhythmic movements and discrete tasks, we run an experiment with a robotic device while subjects bounced a ball in a virtual-reality paradigm in different virtual gravity environments. We showed that the most effective training condition depended on the degree of rhythmicity: Haptic guidance hampered learning of continuous rhythmic tasks, but it promoted learning for repetitive tasks that resemble discrete movements.
Based on the previous results, a new motor learning study with a more complex locomotor task was performed using only the training strategies expected to enhance learning: i) No Guidance, ii) Error amplification, and iii) Noise force disturbance. The motor learning experiment was performed with 34 healthy subjects inside the MRI scanner in order to evaluate the brain activity related to training with the different strategies. Subjects were requested to actively coordinate their legs in a desired gait-like pattern in order to track a Lissajous figure presented on a visual display. The pattern was achieved by moving the knees up and down following sinusoids of equal frequency, but different amplitudes and with a diphase between the left and right legs of 60o. Preliminary analysis from 23 subjects revealed that training without perturbations was especially suitable for a subset of initially less-skilled subjects, while error amplification seemed to benefit more skilled subjects. Training with error amplification, however, limited transfer of learning. Random disturbing forces benefited learning and promoted transfer in all subjects, probably because it increased attention. These results thereby, proved that learning a locomotor task can be optimized when errors are randomly evoked or amplified based on subjects’ initial skill level, thereby confirming our hypothesis. A first preliminary inspection of the brain images showed that training with the different strategies led to the activation of brain networks that are consistent with the areas previously associated with stepping and cycling of the lower limbs (i.e. bilateral activation of the paracentral lobule, the supplementary motor area as well as the cerebellum). We found more widespread bilateral motor network activation during random noise, which could be associated with the augmented complexity of this training condition. Nevertheless, a more exhaustive analysis of the brain images will be performed in order to study the effect that the behavioral data (i.e. the tracking error) might have on brain activation.
This research has the potential to impact on fields other than motor learning, such as robotic rehabilitation. Understanding the underlying mechanisms of motor learning during robotic training is important in order to improve the efficacy of robotic rehabilitation with patients. The hypothesis is that the training strategies that maximize motor learning in less skilled healthy subjects might improve rehabilitation outcomes in more neurologic disabled patients, while patients suffering from mild disabilities might improve recovery when trained with strategies that benefit learning in more skilled subjects. The results from studying the particular brain regions involved in learning might help tailoring motor training conditions to the anatomical location of a focal brain insult. Therefore results from this research project might have a significant impact, increasing the quality of life of many subjects with movement disorders such as stroke survivors. Subjects who have to learn complex motor tasks, such as aircraft pilots and surgeons, could also positively benefit from this research. For example, novel surgeons could be instructed through a similar robotic automatic training strategy to improve their skills while avoiding the danger and high cost associated to training in real situations.
The proposed research project aimed to develop control algorithms in order to enhance motor learning and neural plasticity in robot-aided gait training. The hypothesis behind the research project is that learning a locomotor task can be optimized when errors are decreased or amplified based on subjects’ initial skill level. Therefore, the first project’s objective consisted in performing an evaluation of the effect that the individual subjects’ performance might have on the effectiveness of different forms of robotic training algorithms that amplify or reduce movement errors on motor learning of a locomotor task. The goal of robotic therapy is to develop robotic devices in order to perform exercises which provoke motor plasticity. However, it is still an open question how different rehabilitation strategies contribute to brain restorative processes. Therefore, the second objective of the research project consisted in an evaluation of the brain regions involved in learning when training with different forms of guidance, by means of performing functional Magnetic Resonance Imaging (fMRI) while training with an fMRI compatible walking robotic device.
The first step towards the project completion consisted in developing new control strategies to improve motor learning of locomotor tasks using the MRI-compatible robotic device developed in the host institution (MARCOS). Five different training strategies that reduce or amplify movements errors during training were developed: i) Haptic guidance: errors were eliminated by passively moving the limbs, ii) No guidance: no robot disturbances nor robotic guidance were presented, iii) Error amplification: existing errors were amplified with repulsive forces, iv) Noise disturbance: errors were evoked intentionally with a randomly-varying force disturbance on top of the no guidance strategy, and v) Assist-as-needed: a compliant controller capable to adapt the robot assistance step by step until the value of the assistance converges into an amount slightly lower than the least amount of assistance the subject needs to achieve the desired movement. The new controllers’ performance was experimentally tested inside and outside the MRI scanner and they were proved to provide accurate pneumatic control in MR environments in order to allow assessments of brain activation.
A first motor learning study with 22 healthy subjects was run in order to investigate the effect that the subjects’ individual skill level had on the effectiveness of the designed training modes that amplify or reduce errors on motor learning of a simple locomotor task. Subjects were instructed to relax the dominant leg while it was moved by the robot, and to synchronize the non-dominant leg with the passive leg to achieve a gait-like alternating movement with same amplitude and frequency. A within-subject cross-over design was used to test the effects of training with four different training modes: i) Haptic guidance, ii) No Guidance, iii) Error amplification, and iv) Noise force disturbance. Results indicated that training strategies that reduce or do not amplify errors limit muscle activation during training and result in poor learning gains. Adding random disturbing forces during training seems to increase attention, and therefore improve motor learning. Error amplification was the most difficult strategy for the subjects, aroused the highest muscle activation, brought an after effect and also seemed to provoke the best motor learning in less skilled subjects, perhaps because subjects could better detect their errors and correct them. Skilled subjects could not improve their performance, probably because the task was too easy.
Results from this motor experiment contradicted recent studies that suggested that training with appropriately designed haptic guidance can enhance learning of some motor skills, such as time-critical tasks. Walking is a time-critical task, since it involves a certain level of rhythmicity, and therefore a positive effect of haptic guidance was expected. However, no studies to date have systematically evaluated the effect that haptic guidance have in learning to perform rhythmic tasks, compared to discrete movements (i.e. tasks that include well-defined postures at the beginning and the end of the movement, such as target-oriented reaching movements). In order to investigate the effect of haptic guidance (and no guidance) on learning time-critical continuous rhythmic movements and discrete tasks, we run an experiment with a robotic device while subjects bounced a ball in a virtual-reality paradigm in different virtual gravity environments. We showed that the most effective training condition depended on the degree of rhythmicity: Haptic guidance hampered learning of continuous rhythmic tasks, but it promoted learning for repetitive tasks that resemble discrete movements.
Based on the previous results, a new motor learning study with a more complex locomotor task was performed using only the training strategies expected to enhance learning: i) No Guidance, ii) Error amplification, and iii) Noise force disturbance. The motor learning experiment was performed with 34 healthy subjects inside the MRI scanner in order to evaluate the brain activity related to training with the different strategies. Subjects were requested to actively coordinate their legs in a desired gait-like pattern in order to track a Lissajous figure presented on a visual display. The pattern was achieved by moving the knees up and down following sinusoids of equal frequency, but different amplitudes and with a diphase between the left and right legs of 60o. Preliminary analysis from 23 subjects revealed that training without perturbations was especially suitable for a subset of initially less-skilled subjects, while error amplification seemed to benefit more skilled subjects. Training with error amplification, however, limited transfer of learning. Random disturbing forces benefited learning and promoted transfer in all subjects, probably because it increased attention. These results thereby, proved that learning a locomotor task can be optimized when errors are randomly evoked or amplified based on subjects’ initial skill level, thereby confirming our hypothesis. A first preliminary inspection of the brain images showed that training with the different strategies led to the activation of brain networks that are consistent with the areas previously associated with stepping and cycling of the lower limbs (i.e. bilateral activation of the paracentral lobule, the supplementary motor area as well as the cerebellum). We found more widespread bilateral motor network activation during random noise, which could be associated with the augmented complexity of this training condition. Nevertheless, a more exhaustive analysis of the brain images will be performed in order to study the effect that the behavioral data (i.e. the tracking error) might have on brain activation.
This research has the potential to impact on fields other than motor learning, such as robotic rehabilitation. Understanding the underlying mechanisms of motor learning during robotic training is important in order to improve the efficacy of robotic rehabilitation with patients. The hypothesis is that the training strategies that maximize motor learning in less skilled healthy subjects might improve rehabilitation outcomes in more neurologic disabled patients, while patients suffering from mild disabilities might improve recovery when trained with strategies that benefit learning in more skilled subjects. The results from studying the particular brain regions involved in learning might help tailoring motor training conditions to the anatomical location of a focal brain insult. Therefore results from this research project might have a significant impact, increasing the quality of life of many subjects with movement disorders such as stroke survivors. Subjects who have to learn complex motor tasks, such as aircraft pilots and surgeons, could also positively benefit from this research. For example, novel surgeons could be instructed through a similar robotic automatic training strategy to improve their skills while avoiding the danger and high cost associated to training in real situations.