| Brain machine interfaces (BMI) for the restoration of arm movements promise to dramatically improve the quality of life of tetraplegic or amputation patients. Over the past 20 years several groups have developed and tested techniques, in both monkeys and humans, to record neural commands, process them, and use them for the control of computer cursors and robotic arms. Several issues persist, some of which I tried to address and present in this thesis.;An important problem with current BMIs is the tradeoff between movement speed and stability. Given the limited bandwidth of the recorded signals, which come from only about 100 neurons in motor cortex, it is hard to achieve the levels of performance of an intact arm using a BMI. I was part of a team that created a dual-state BMI, capable of decoding fast and slow movements independently. The outputs were combined automatically by a classifier, allowing for a seamless transition between the two states. The performance of our BMI was superior when compared to a regular, single filter BMI.;Restoration of movement is only half of the goal after a spinal cord injury. Sensation below the level of injury is lost, which damages the patients' ability to know the state of their body. I designed and carried out experiments that allowed me to assess whether sensory information can be conveyed by stimulating the primary somatosensory cortex (S1) of awake, behaving monkeys. We found that by electrically stimulating multiple sites in S1, we can increase the probability of stimulus detection while minimizing the current applied at any given site.;The third problem I investigated was the inability to control interaction forces in current BMIs. These forces will arise whenever the controlled effector, be it a robotic arm or the patient's paralyzed arm, is used to interact with the environment. The solution I implemented consisted of decoding the activity of virtual muscles from primary motor cortex (M1) and using that signal to drive a virtual arm that comprised a simple musculoskeletal model. By modulating the co-activation of antagonistic muscles it was possible to change the impedance of the virtual arm, allowing the user not only to move the arm but also modulate the way it responds to external perturbations. |
