Abstract
This paper presents a method for a robotic meal support system to handle tofu, a non-rigid food item. The system utilizes a laser range finder (LRF) and a 6-degree-of-freedom manipulator equipped with a spoon. By measuring 3D surface coordinates and determining optimal scooping positions, the robot can handle tofu without requiring pre-meal cutting by a caregiver. Preliminary experiments demonstrate the effectiveness of this autonomous approach in handling fragile, non-rigid objects.
Introduction
For individuals with upper-limb disabilities, independent eating is a significant challenge. While commercial meal assistance robots exist, they often require food to be pre-cut into uniform, bite-sized pieces. This preparation burden limits user autonomy and adds to the caregiver's workload. This research addresses the challenge of handling "soft" or non-rigid objects—specifically tofu—which is common in care-giving diets but difficult for traditional robots to handle without crushing.
System Methodology
The proposed system integrates a stationary Laser Range Finder (LRF) with a 6-DOF robotic manipulator. The operational cycle follows a structured "sense-plan-act" workflow to adapt to the changing shape of the food as it is consumed.
3D Perception and Extraction
The LRF captures a 3D point cloud of the plate. To isolate the tofu, the system performs "subtraction processing," comparing the current scan to a pre-recorded scan of the empty plate. This computationally efficient method allows the system to identify the tofu's volume and boundaries regardless of its initial shape or orientation.
Surface Segmentation and Trajectory Planning
The isolated point cloud is segmented into "top" and "side" surfaces. The system then determines the optimal insertion point for the spoon based on the tofu's edges. Rather than a linear movement, the robot executes an "osculating circle" trajectory—a circular arc path that mimics human scooping. This geometric approach ensures the spoon enters and exits the food at angles that minimize structural failure, preventing the tofu from crumbling.
Experimental Results
The system was validated using a standard block of tofu (6cm x 6cm x 5.5cm). In preliminary tests, the robot successfully performed five consecutive scooping iterations, clearing the majority of the block. By re-scanning the plate after each action, the system dynamically updated its model, effectively treating the deforming tofu as a "new" object each time. This iterative sensing bypasses the need for complex physics simulations of non-rigid body deformation.
Conclusion
The research demonstrates a successful proof-of-concept for an autonomous meal support system capable of handling uncut, soft food. Future work will focus on refining the algorithm to account for plate geometry and solving the "last mile" problem, where small pieces of tofu slide across the plate. Potential strategies include using the plate's rim as a backstop to provide the necessary resistance for final scoops.
Summary updated