Research

Balance

  • Current Research
  • Previous Research

Overview

Advanced Displays for Efficient Training and Operation of Robotic Systems

Principal Investigator:
Charles M. Oman, Ph.D.

Organization:
Massachusetts Institute of Technology

In an effort to improve training efficiency, Dr. Charles Oman and colleagues are developing methods for assessing astronaut skill in controlling space telerobotic systems, and improving the efficiency of training. The team has tested astronaut spatial abilities and found they are predictive of high performance during early training. Other experiments are conducted on non-astronaut subjects training on a robotics workstation simulator at MIT. So far, the team’s research has demonstrated that spatial abilities correlate with several measures of performance in training as a primary or secondary robotics operator. The group proposed several changes in robotics aptitude evaluation methods which have been implemented at NASA. The group is also examining the effects of bimanual control skills and handedness, and developing and testing better visual interfaces for future space and lunar surface operations.

NASA Taskbook Entry


Technical Summary

The long-term objectives of this four-year project address three specific aims related to human performance during space telerobotics training. We are collaborating with the NASA Johnson Space Center (JSC) Robotic Systems Training Group. The project is in its third year.

Astronaut robotics trainees vary significantly in their initial performance, ability, learning rate and level of mastery. Because the process of training astronauts to be qualified robotics operators is so long and expensive, NASA needs tools to predict performance and customize training. Our scientific goal has been to understand how individual differences in spatial and manual control abilities impact learning and performance.

Specific Aims

  1. Astronaut candidates currently take an Aptitude for Robotics Test (ART), and those selected proceed to Generic Robotics Training (GRT). Using a logistic modeling approach, we investigated how well an astronaut's ART scores and an additional set of mental rotation, perspective taking and visualization tests predicted spatial performance in subsequent training. We found ART was not a reliable predictor and proposed changes in ART metrics to improve the predictive power. These changes were implemented and used in the current round of astronaut testing and GRT training. During the coming year, we plan to re-evaluate ART using this new data. Logistic regression analysis of mental rotation and visualization scores allows us to predict who will achieve a top score in qualification evaluations, but not those who fail (partly because very few do). Model predictions are reliable enough to use in customization of regular and remedial training, but not to make career-defining decisions. Additional GRT and spatial ability data is also being obtained for analysis this year in collaboration with the JSC Robotics Training Branch.
  2. Our second objective has been to study performance and learning in a controlled laboratory setting using a space telerobotics training simulator at Massachusetts Institute of Technology (MIT). The simulator recreates the NASA Basic Operational Robotic Instruction System (BORIS), used in NASA's introductory Generic Robotics Training course (GRT), and also the International Space Station (ISS) environment. In a series of three previous experiments, we consistently found that a trainee's early performance and learning in relatively simple GRT-like "fly-to" and pre-grapple tasks correlates with their spatial abilities. We believe this is because mental rotation and visualization abilities are important for integrating the multiple video camera views used when performing robotics tasks.

    This year, we completed two more experiments. In our prior research, camera configurations were controlled by the experimenter. In reality, cameras are selected by the primary or secondary operator. In the first experiment (n=21), we found that while acting as a secondary operator, camera selection performance and ability to identify arm clearance issues was also correlated with the individual subject's (Vandenberg) mental rotation, (Purdue Spatial) visualization, and (Kozhevnikov 2D) perspective-taking ability.

    A second experiment (n=20) studied the effects of spatial ability, handedness and joystick configuration on "fly-to" performance requiring multi-axis movements. Like spacecraft, the space shuttle and ISS robotic arms are always controlled using a three degrees of freedom (DOF) rotational controller in the right hand, and a three DOF translational controller on the left. Current hand preference theories (e.g., Guiard) suggest that right-handed astronauts should be at a particular advantage with this physical hand controller arrangement. As in our prior experiments, we found that spatial ability scores predicted task performance. However, to our surprise, we found no large or consistent effect of handedness (Edinburgh questionnaire), or whether the rotational hand controller was in the dominant or non-dominant hand, even in the early stage of training. As when using an "Etch-A-Sketch," robotics operators must learn to parse different spatial axes to different hands. Happily for left-handed astronauts, right-handed operators apparently have no significant advantage in this regard.
  3. So far, our MIT experiments have studied performance only during the first 1-2 days of training while operators perform relatively simple fly-to, camera selection and clearance monitoring tasks. Are spatial abilities as important during the later phases of GRT and more advanced training? Based on discussions with the Robotics Training Branch, this year we developed a new series of training protocols and performance metrics in order to study performance during more advanced tasks. New experimental scenarios developed include grapple, loaded-fly-to, auto sequencing and free-floating payload track-and-capture. With the impending retirement of the shuttle, astronauts must be able to capture a drifting logistics supply vehicle.

    We have also developed a side task to assess workload. Trainees are able to perform complete sequences of robotic tasks, and the more realistic protocols will allow us to study the process of complex skill acquisition under multi-day advanced training scenarios and quantify spatial ability effects. Also, in a companion study being initiated this year, we plan to assess the effects of fatigue and sleep deprivation.
     

Earth Applications

Our goal is to improve the efficiency of robotic training via improvement of current pedagogies and development of new teaching tools. Improved training methods provide a framework for designing future in-flight training procedures during long-duration missions. The project will also demonstrate how individual differences in spatial and manual control skills affect performance of critical operational skills, including complex robotics tasks associated with post-space shuttle era ISS operations.

This project's funding ended in 2011