Progressive Haptic Guidance for Training in a Virtual Dynamic Task

The implementation of training virtual environments (TVEs) is intended to reduce risk, improve and accelerate learning over traditional training methods, thereby transferring what is learned in the simulation to the targeted real world task. One type of TVE employs a type of robotic force feedback, also called haptic guidance, to assist the human trainee in performing the critical components of the task. Prior work suggests that these haptic guidance schemes perform best when the level of guidance is based on the trainee's changing level of performance during training. Our objective is to demonstrate that expert based progressive haptic guidance can accelerate and improve training outcomes over visual or practice-only methods. To that end, we design a guidance scheme based on a detailed analysis of performance differences between expert and novice trainees. The guidance design is then tested with two trainees in a dynamic task experiment thereby verifying its functionality.

We employ a virtual environment with haptic feedback that simulates a dynamic target-hitting task. This TVE was previously developed by Li et al. to study the efficacy of an error reducing guidance scheme. However, we extend the possible assistance modes available by introducing progressive guidance that is based on expertise. Both the task and the proposed guidance are shown in Fig. 1. In previous work, we analyzed the performance of experts and novices during task execution and discovered two independent components required to successfully complete the task. The spatial component is to keep the input joystick, and therefore the output disc, on the trajectory axis as much as possible so as to ensure target acquisition. The temporal component is to excite the input joystick at the resonant frequency of the virtual system dynamics so that the disc will oscillate rapidly and with sufficient amplitude between the targets. In order to measure how well the trainee is performing in these two components, we introduced two measures: etraj (trajectory error) and finput (input frequency). The etraj is the trial sum of the absolute errors in position, while the finput is computed via a fast Fourier transform of the position data along the trajectory axis to determine the amplitude and frequency of the input motion being applied to the system. High correlation coefficients were statistically verified between these two measures and the hit count measure. The etraj and finput are then used as the inputs to the haptic guidance algorithm and a similar visual guidance algorithm.

The guidance proposed in this work is comprised of two orthogonal regions as shown in Fig. 1 in order to demonstrate the two task components. The first region (shown in dark gray) indicates the maximum allowable deviation from the trajectory axis that will still result in a target acquisition, thereby reducing etraj. The second region (shown in light gray) oscillates at the resonant frequency of the second order system and with an amplitude that will, if tracked, ensure sufficient output amplitude at a frequency near the resonant frequency to acquire the maximum number of targets. For the visual scheme, these two regions are represented by colored bars whose intensities diminish as performance improves in each of the two measures. Similarly, in the haptic scheme, the edges of the regions are represented by stiff virtual walls. The minimum force required to penetrate the walls is progressively reduced as performance improves thus gradually shifting primary control from the robot to the trainee as the training protocol progresses. Both the visual and haptic guidance employ an exponentially decaying gain. When three successive trials show improvement in performance, the gain decreases. In contrast, when three trials show degrading performance, the gain increases.  Fluctuating performance trends cause the gain to remain unchanged.

We contend that haptic guidance schemes for virtual environment training must be based on the task components that are critical to successful task completion. Minimization of trajectory error and excitation of the virtual dynamic system near resonance were identified from expert performance as key components for this task.  Our guidance design will allow comparison of a haptic guidance scheme to traditional visual and practice-only schemes.  Pilot results indicate that the progressive guidance design is successful for both visual and haptic guidance