Force Distribution Sensor Based on Externally Observable Three-Dimensional Shape Deformation Information

Recent developments in segmentation and inexpensive RGBD cameras make it possible to easily perform object shape extraction. This enables a new principle for externally observable force sensors, in contrast to traditional force detection that relies on contact-based displacement measurements. Among the sensors that estimate force from visual inputs, those using machine learning methods based on experimentally generated data have the possibility to outperform other visual sensors. However, the existing models have very limited applicability due to their low adaptability to environmental changes and their capacity for only single-point estimation. Therefore, in this letter, the authors propose and verify a method that can flexibly respond to environmental changes and estimate force at multiple points in three dimensions. The authors conducted a demonstration experiment and confirmed that estimation was possible with an error of approximately 20%. Observation-type force sensors are expected to enable a variety of applications that are difficult to achieve with conventional sensors, such as sensing people in living environments and the development of sensing robots.