Abstract
This paper introduces a novel framework for human-robot handover utilizing a low-cost, vision-based tactile sensor. By monitoring the physical displacement of a marker within a flexible sensor "skin," the system intelligently determines when to release an object to a human receiver. The approach proves effective even for fragile or irregularly shaped items, achieving a 94% success rate in experimental trials.
Introduction
Successful human-robot handover requires precise coordination. Releasing an object too early results in drops, while releasing too late causes a "tug-of-war" sensation that diminishes user experience. Traditional solutions often rely on external cameras, which are prone to occlusion, or expensive force-torque sensors. This research addresses these challenges by using internal tactile feedback to sense a human's intent to take an object, specifically focusing on the transition from robot grip to human control for a diverse range of objects, including soft and deformable items.
Methodology
The core of the system is a custom tactile sensor comprising a latex rubber layer, an internal tracking marker, a camera, and a pressure-controlled chamber. The software architecture is divided into three distinct modules to process visual data into a release command.
- Segmentation Module: This module utilizes OpenCV to isolate the tracking marker from the camera feed, producing a high-contrast grayscale image for analysis.
- Dot Tracking Module: It monitors the coordinates of the marker's center. As a human pulls the object, the resulting deformation of the latex skin shifts the marker's position.
- Decision Module: This module calculates the accumulated displacement ($M$) of the marker over a specific time window. If $M$ exceeds a predefined sensitivity threshold ($S$), the system triggers the robot's gripper to open.
The researchers evaluated nine combinations of sensitivity ($S$) and internal pressure ($P$) to optimize for both user comfort and object safety.
Experimental Evaluation
The framework was tested through an exploratory user study involving 10 participants. The study utilized eight objects selected for their varying stiffness and geometry: paper, fabrics, a cup, dough, strawberries, raspberries, a cable, and a prism.
Results and Discussion
The system demonstrated robust performance with an overall success rate of 94%. Rigid and flat objects, such as paper and cups, saw near-perfect success rates (99-100%). However, performance slightly decreased for complex items like cables (84%) and raspberries (90%). The cable's small diameter occasionally failed to deform the skin enough to trigger the central marker.
Key insights from the study include:
- Displacement as a Force Proxy: The system successfully uses marker displacement to "feel" human pull without requiring explicit force calculations in Newtons.
- User Preference: Participants reported the highest "fluency" and comfort when the sensitivity was at its maximum ($S = 0.001$), indicating a preference for highly responsive robotic behavior.
- Pressure Trade-offs: While higher internal pressure ($P$) stabilizes the grip, it increases the risk of damaging fragile items. Sensing times below 500ms were found to be ideal for perceived comfort.
Conclusion and Future Work
The study confirms that low-cost optical tactile sensing is a viable path for collaborative robotics. To address current limitations, such as the failure to detect off-center grasps on thin objects, future work will focus on implementing multi-marker arrays for better spatial resolution. Additionally, the researchers plan to develop dynamic pressure control systems that automatically adjust based on the detected stiffness of the grasped object.
Summary updated