Bi-modal hemispherical sensors for three axis force and contact angle measurement

Abstract

Humans and animals demonstrate a unique ability to interact with the physical environment around them with coordination and control, including moving quickly across rugged terrain or deftly handling small objects. Much of this success is related to our ability to accurately perceive the world around us through a sense of touch. In order to better perform dynamic, physical interactions, such as locomotion or manipulation, robots need to be able to accurately measure contact locations and forces. However, many existing sensors do not satisfy the stringent requirements or do not supply sufficient information for robotic locomotion and manipulation. This thesis builds on work by a previous PhD student, Meng Yee (Michael) Chuah, of the MIT Biomimetic Robotics Laboratory to develop stress field based force sensors for use in robotic applications. The concept of stress field based force sensing consists of pressure sensors embedded within a rubber hemisphere.

The pressure sensors sample the stress distribution within the rubber, and use these signals to reconstruct the applied force. This type of sensor is inherently robust, low cost, and insensitive to inertial noise. This work focuses on the development of bi-modal hemishperical sensing technology for two novel sensor designs --

one footpad sensor intended for use in a high force range corresponding to legged locomotion, and one fingertip sensor intended for use in a lower force, higher sensitivity range, corresponding to robotic manipulation applications. Both sensors have the ability to simultaneously measure applied force in three axes (Fx, Fy, Fz) and the contact location of quasi-point contact, as described by two angles ([theta], [phi]) in real time (1kHz and 200Hz, respectively). The sensors each contain eight embedded pressure sensors, and can accurately reconstruct the five desired outputs from the eight input signals using either a Gaussian process regression (GPR) estimator, or an artificial neural net (ANN). The performance of both sensors using each estimator is quantified through testing on multiple data types. The properties of the sensor, including sensitivity, repeatability, and drift over time are also characterized.

The performance of the footpad sensor is further validated through two applications on a robotic arm. In the first, normal force and contact angle information from the footpad sensor is used to accurately sense and track a moving surface. In the second, normal and shear force information from the sensor is used to detect and prevent slip. Overall, these sensors demonstrate the ability to quickly and accurately measure forces in three axes and contact surface normal, while being robust and low cost. These sensors have the potential to greatly improve the capability of robots to perform dynamic, physical interactions with the world around them.