Robotic devices have emerged as promising solutions to support motor training in physical rehabilitation, surgery, and sports, but the optimal control strategies for robotic motor training are still unclear. This dissertation focuses on the development and evaluation of algorithms that modulate the challenge level experienced by a trainee in order to optimize training. Challenge level has been proposed to affect both motor learning and motivation, but these effects, in the context of robotic-based training, are not well understood. To understand these effects, we developed three novel experimental paradigms and studied motor learning of a complex motor skill in humans, and motor rehabilitation after spinal cord injury in a rat and in a non-human primate model.
The first experiment focused on motor learning in humans without impairment. We modulated the difficulty of a virtual golf task by either reducing or augmenting kinematic errors with a haptic robot while participants putted to two target locations. We found that the training conditions had mixed effects on learning, but signi1cantly affected participants' subjective experiences of training. Specifically, robotically reducing errors improved self-reports of competence and satisfaction, while augmenting errors worsened reports. These effects persisted days after the robotic manipulations ceased, even when participants' performance returned to normal. These results indicate that robotic training can modulate, with lasting effects, the subjective experience of training.
The second experiment focused on hand motor rehabilitation after spinal cord injury in a rat model. We modulated the difficulty of a pulling task by controlling the force level required to achieve success. Rats trained either with constant, low forces, or with an adaptive algorithm that controlled the success rate to be 50%. We found that animals in the low-force group attempted more pulls and were more successful at achieving the task than animals in the adaptive group. However, animals in the adaptive group recovered more grip pulling strength, as evidenced by higher pulling forces in the assessments conducted throughout the rehabilitation process. Thus, the benefits of training at an adaptively-controlled, high-challenge level exceeded the benefits of increased numbers of practice repetitions achieved at a lower challenge level.
The third experiment focused on hand motor rehabilitation after spinal cord injury in a non-human primate model. We developed a robotic device that provides in-cage, self-training exercises for animals with a spinal cord injury that affects their ability to use the right hand. The device uses training algorithms that accommodate animals with a wide range of capabilities to ensure they can engage in training even after suffering a severe injury. We found that animals were able to train with the device both before and after the lesion by setting the appropriate difficulty level.
These findings lead us to propose the basis for a framework to understand the effect of challenge on performance gains and motivation during unsupervised learning for robotic-based motor training. This framework proposes that in environments where the trainee chooses training amounts, total learning is determined by the product of two component relations: 1) the challenge level and training frequency, and 2) the challenge level and learning-gain per trained movement. Since the challenge level has opposite effects on these component relations, this product yields a parabolic relation between the challenge level and the learning-gain per training frequency. Thus we can find a challenge level where total learning is maximum. The challenge level corresponding to this maximum represents the optimal challenge level for unsupervised training. Robotic training devices provide a means to both identify this optimal challenge point and control the challenge level so that training can be carried out around that point.