Whole body humanoid robot force-sensing for compliant interaction with the environment
Guardat en:
| Autor: | |
|---|---|
| Data de publicació: | 2026 |
| Descripció: | This thesis proposes a method for integrating a system capable of sensing external forces applied to the mobile platform and torso of a humanoid robot. The humanoid robot is characterized by its approximate weight of 440 kg and by the use of mecanum wheels, which provide omnidirectional mobility, allowing it to move easily in dynamic environments where it interacts with people. Force sensing, together with force control strategies such as impedance and admittance control, is essential for enabling robots to interact safely with people and objects. These technologies provide robot mechanisms with compliant behavior during physical interactions and enable capabilities such as the manipulation of large and heavy objects through physical contact, as well as the communication of human intentions. Despite their importance, force sensing systems are still not widely integrated into robotic platforms, including many research robots. The development and integration of the force-sensing system presented in this work involve several significant challenges. These include the design and fabrication of a multi-degreeof-freedom force-torque sensor with acceptable mechanical crosstalk; the precise alignment and installation of the sensing elements, which in this work are strain gauges; the design of a force-torque sensor with sufficient sensitivity for the intended application while maintaining a high load capacity to withstand the mechanical stresses produced by the weight of a 450 kg humanoid robot without reaching the material yield strength; the design of data acquisition and communication systems capable of reducing the electromagnetic interference generated by the robot’s high-power body components while achieving data transmission rates sufficient for real-time force control strategies, such as impedance and admittance control; the in situ calibration of the force system to compensate for the loads introduced by integrating the force-torque sensors into the robot structure; and the development of an appropriate strategy to address the nonlinearities of the complete force sensing system. In this work, these nonlinearities were modeled using a One-Dimensional Convolutional Neural Network with residual connections. Until now, few robots have integrated force sensing into mecanum-wheel mobile platforms. In this thesis, we address the problem of implementing force sensing in a heavy mecanum-wheel robot while maintaining enough force sensitivity. Such system presents several challenges. One of the main challenges is that the calibration must be performed in situ, which is particularly difficult if a static calibration procedure is intended because of the limited available space and the absence of suitable calibration equipment that can be attached to both the robot and the sensors simultaneously. Another major challenge is the estimation of external forces and contact locations while the mechanical coupling between the force-torque sensors, the mobile platform, and the torso is considered. This coupling introduces nonlinear effects caused by friction, backlash, and structural elasticity in the mechanical connections between the different robot components, making the force estimation problem considerably more complex. The force sensing system developed in this thesis consists of low-cost force-torque sensors and data acquisition and communication systems, which serve as the basis for the implementation of a one-dimensional convolutional neural network with residual connections. This neural network estimates the equivalent force applied to the torso up to a maximum of 30 N and to the mobile platform up to a maximum of 100 N. Furthermore, it estimates the three-dimensional coordinates of the equivalent contact point once an approximate threshold of 5 N is exceeded for forces applied to the torso and 20 N for forces applied to the mobile platform. The neural network algorithm was able to distinguish useful information from noise through the implementation of a dynamic in situ calibration, which enabled the modeling of the nonlinearities introduced by friction and backlash present in the mechanical transmissions, as well as by the coupling of the force-torque sensors to the mobile platform, the coupling of the mobile platform to the torso of the humanoid robot, and the elasticities present between the different parts of the robot. The experiments conducted with the force sensing system demonstrated satisfactory results in both the estimation of external forces parallel to the robot’s plane of motion and the estimation of the contact point. Consequently, this thesis proposes a method that can serve as a starting point for the implementation of admittance or impedance control systems. Such systems could enable, in the future, the manipulation of large and heavy objects using the mobile platform of a heavy humanoid robot, as well as the interpretation of commands conveyed through physical contacts applied to the robot’s torso. |
| Pais: | Kérwá |
| Institution: | Universidad de Costa Rica |
| Repositorio: | Kérwá |
| Idioma: | Inglés |
| OAI Identifier: | oai:kerwa.ucr.ac.cr:10669/105194 |
| Accés en línia: | https://hdl.handle.net/10669/105194 |
| Paraula clau: | WHEELED HUMANOID ROBOT COMPLIANT INTERACTION WHOLE BODY HUMANOID ROBOT FORCE-SENSING |