In human-machine collaborative systems, robot joint compliance and human-input dynamics lead to involuntary tool motion into undesired regions. To correct this, a set of methods, called dynamically-defined virtual fixtures, was previously proposed to create a movable virtual fixture that stops the user at a safe distance outside the forbidden region. In this work, a new method, called the force-based method, was added. A vision system was introduced for real-time tool tracking. Additionally, we implemented a closed-loop controller with the virtual fixtures that allows the user to reach, but not enter, the forbidden region. Two user experiments were conducted on a 1-DOF testbed to evaluate the virtual fixture methods. The first experiment showed the effectiveness of the virtual fixtures in preventing the penetration. However, the absence of haptic feedback in the closed-loop implementation resulted in boundary penetration. In the second experiment, visual feedback was used to compensate for the lack of haptic feedback. User cognitive load was added as an inhibiting factor in a human-machine cooperative setting. The experiment showed a significant reduction in penetration with visual feedback, while the addition of cognitive load did not significantly increase the penetration.