The Use of Gaze-Derived Confidence of Inferred Operator Intent in Adjusting Safety-Conscious Haptic Assistance

TL;DR

The paper addresses the problem of teleoperation under hazardous conditions by restoring a level of natural control through gaze-driven intent inference that informs haptic assistance. A naive Bayes-based intent predictor uses three gaze features to compute a posterior probability $p(i|G_1,G_2,G_3)$, which is scaled to a confidence score $c_i$ (with $c_i \ge 0.5$ indicating intent), and is used to modulate a Gaussian potential-field fusion that blends joystick input with the gaze-derived target to generate a final command $c$ while applying a safety boundary around the target. The two key haptic fixtures—a guidance force and a safety boundary—are adjusted in real time by $c_i$, leveraging a Gaussian potential field with covariance $\Sigma$ to handle target uncertainty. Parameter design for the safety boundary (radius $S$, height $H$, cone angle $\theta$) balances risk and maneuverability, with eight sets evaluated to minimize task failure. Experimental results on cutting and grasping tasks show improved performance with haptic assistance and confidence-based adjustments, though effectiveness is task-dependent and not always statistically significant; the work demonstrates practical gains in safety, accuracy, and user comfort in teleoperation and points to future enhancements via richer intent signals and contextual gaze analysis.

Abstract

Humans directly completing tasks in dangerous or hazardous conditions is not always possible where these tasks are increasingly be performed remotely by teleoperated robots. However, teleoperation is difficult since the operator feels a disconnect with the robot caused by missing feedback from several senses, including touch, and the lack of depth in the video feedback presented to the operator. To overcome this problem, the proposed system actively infers the operator's intent and provides assistance based on the predicted intent. Furthermore, a novel method of calculating confidence in the inferred intent modifies the human-in-the-loop control. The operator's gaze is employed to intuitively indicate the target before the manipulation with the robot begins. A potential field method is used to provide a guiding force towards the intended target, and a safety boundary reduces risk of damage. Modifying these assistances based on the confidence level in the operator's intent makes the control more natural, and gives the robot an intuitive understanding of its human master. Initial validation results show the ability of the system to improve accuracy, execution time, and reduce operator error.

Figures (16)

• Figure 1: System overview
• Figure 2: Overall control flow which illustrates the role of the operator's gaze in the hybrid joystick-gaze control method. The blue lines represent the control flow for traditional teleoperation.
• Figure 3: Example of the gaze features used to determine the intent confidence. X and Y are the screen coordinates in pixels.
• Figure 4: An illustration of the potential hybrid control method for a two-dimensional control space. $p_g$ is the goal point, $p_j$ is the joystick point, $c$ is the combined point, and $\sigma$ is the size of the field in each direction.
• Figure 5: Magnitude of the guidance force in two dimensions for a combination with equal strengths in both $X$ and $Y$. The target position is placed at (0.5, 0.5). The profile of the force is the same for all directions of approach.