Abstract
Sense of touch is an essential sensing method for humans to interact with the unstructured and dynamic environment, especially in scenarios where the vision is restricted or unavailable. However, it remains an open challenge for robots to do so as effective as humans. To advance the technology and understanding in robotic haptic interaction and perception, the following novel algorithms have been proposed and validated experimentally in this thesis.Firstly, according to surveyed knowledge, precise contact force and explore velocity control during haptic exploration remained highly unexplored, especially on objects with dynamic movement. To tackle this challenge, a novel control algorithm utilizing intrinsic contact sensing (ICS) information and fuzzy control logic is created. A contact sensing finger has been created and utilized to test the control algorithm in real-time on nonstationary objects with different shapes, stiffness, surface friction, roughness and patterns. Another experimental platform utilizing a robot arm and force-torque sensor has been designed for validating the algorithm in larger working space and force ranges. With both platforms, the proposed algorithm showed stratified control results with distinctive mean square error and standard deviation values.
Secondly, to utilize the physical and geometry properties of the object extracted through haptic surface exploration, a multivariate Gaussian Bayesian classifier is created and validated experimentally. The data of forces and contact locations obtained through haptic sensing during surface exploration are interpreted into surface friction coefficient, roughness and geometry features. With the 3-dimensional features vector of the objects, the classifier has been trained and validated extensively using objects with different physical and geometry properties. The classifier showed an average recognition accuracy of 93% with features obtained through real-time haptic surface explorations.
Thirdly, to bridge the created algorithms, an active haptic perception algorithm has been v created which can decide the next best haptic exploration procedure (EP) for highly accurate and efficient object perception. The algorithm makes decisions based on both the information gain derived through the multivariate Gaussian Bayesian classifiers and the expected time cost consumed by the EPs. Both simulated and experimental validations have been carried out to estimate the algorithm extensively. As a result, the proposed algorithm showed distinctive advantages with much higher accuracy and lower cost compared with the random method.
Furthermore, in collaboration with Pisa University, a comparison of two contact location estimation methods on deformable surfaces based on ICS has been carried out. The first method is a closed-form approach and the other is an iterative approach.The iterative method has been estimated with real world experiments. And a method combining the two methods, using the closed-form approach as the initial guess for the iterative approach is proposed and in the validation experiments, it showed convergence under 1ms with errors lower than 1 mm. Ultimately, a software toolbox containing all the necessary functionalities has been created and shared online, with the support of Pisa University.
| Date of Award | 1 Feb 2020 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Hongbin Liu (Supervisor) & Matthew Howard (Supervisor) |