Document Type
Report
Abstract
In a complex work environment, an operational challenge may be controlling the multitude of independent objects that are required for a manual task and/or object manipulation, such as experienced by astronauts, surgeons, underwater operations, manufacturing etc. One possible option is the enhancement of the person’s capabilities through a robotic supernumerary limb (Figure 1). The idea has already been implemented in a limited application, where the United States Army Research Laboratory developed a passive limb to aid soldiers in weapon carrying and stabilization[1]. Researchers are also creating supernumerary robots. Parietti, et al. [2] designed the Supernumerary Robotics Limbs to address aircraft manufacturing challenges by assisting workers with object positioning and task assistance. Others have used the robotic supernumerary limb as a third arm to simultaneously draw or play instruments in coordination with the natural arms [3]. However, the ability to control an assistive robot arm while simultaneously using both natural arms for tasks has not been well studied. There are several options for a control input to the assistive robot arm, and we are particularly interested in extending the neuromuscular connection beyond the natural limbs and to the robot through surface electromyography (sEMG). Electrical signals produced by muscles are recorded at the skin’s surface and used to communicate the human’s intent. The human learns to produce a particular muscle contraction pattern that corresponds to predetermined action. Prior work in RASCAL (Robotics, Autonomous Systems, and Controls Laboratory) has used sEMG to control a computer cursor [4-6], teleoperate a mobile robot [7], and a simulated prosthetic arm [8, 9]. We wanted to extend our single-site sEMG control method that used a single muscle for 2 commands [6] to 4-6 commands for an assistive robot arm. However, we did not know the effect of increasing the number of commands on subject learning, training time, proficiency, or any interactions factors, such as perceived cognitive workload and trust. Over the course of the Link Fellowship, and accounting for the pandemic-induced changes, the objectives were: 1. Study 1: Evaluate the sEMG control method with increased DOFs prior to significant investment in a more complicated, robotic research testbed through a cursor-to-target task. Study the effects of different training methodologies on performance, perceived cognitive workload, and trust. 2. Study 2: Determine whether applying machine learning algorithms to the control method would aid in subject performance. 3. Study 3: Setup the hardware and software for a virtual Human-Robotics Integration testbed to conduct limited pilot studies. At the time of this report, the first objective has been completed with the other two objectives in progress. In Study 1, we hypothesized that varying training methodologies would affect the subjects’ performance, perceived cognitive workload, and trust in different ways. However, we generally expected all subjects to have similar results by the end of the study when they were sufficiently trained. A more detailed explanation of the hypotheses is available in Ref. [11].
Publication Date
10-2020
Recommended Citation
O'Meara, Sarah M., "Human-Automation Interaction Testbed for Evaluating User Control Methods and Training" (2020). Link Foundation Modeling, Simulation and Training Fellowship Reports. 27.
https://repository.fit.edu/link_modeling/27
Standard cover form for report
Comments
Link Foundation Fellowship for the years 2019-2020.