Abstract Spinal cord injury resulting in paralysis affects more than 250,000 people nationwide with over 12,000 new cases each year. More than half of all SCIs occur at cervical levels resulting in paralysis below the neck with less than 1% achieving full recovery. Paralysis results in a loss of independence, and the need for caregiver assistants to perform activities of daily living. This loss of independence adds to the emotional, financial, and family burden of this chronic condition. Hardware is now available that holds the potential to restore lost arm function after spinal cord injury. Specifically, tiny arrays of implanted intracortical microelectrode are able to detect the neural firing patterns in the areas of the brain responsible for producing reaching movements. Also implanted stimulating electrodes implanted on or near the peripheral nerves are able to reanimate muscles after paralysis. Combining these implanted recording and stimulation technologies holds the potential to restore arm movement controlled by one's own thoughts. However, much work is needed to refine the control algorithms that translate the brain signals into the stimulation patterns needed to make the desired movements. This study will evaluate three methods of improving control algorithms for restoring reaching in both human and animal models. Specifically we are exploring options for controlling not only the motion of the limb, but also how well the limb resists perturbations from external forces. Increasing limb `stiffness' by activating opposing or `antagonist' muscles will help to stiffen and stabilize the limb making it resistant to unintended movement if bumped. In the first method, we are incorporating automated stiffness control based on context (i.e. limb stiffness is modulated based on the intended speed and acceleration/deceleration we decode from the brain signals). In the second method, we will extract a separate `stiffness' command signal from the brain and use that cortically-derived command to adjust limb stiffness in real time. In the third method, we put individual recorded neurons in direct control of the muscle stimulators using a simple linear brain-to-muscle-stimulator mapping. This method puts the brain in charge of learning with practice how to optimize muscle activation to modulate stiffness as needed. Optimizing how stimulation is delivered to control stiffness is important for practical use of this technology because too little stiffness will allow the arm to be easily bumped off course whereas more stiffness than is needed will waste stimulator batteries and can cause muscle fatigue.