Enhancing robotic agents through tactile perception and multi-modal interfaces

传统工业机器人配备了高度专业化的传感、感知和驱动套件，专为重复一系列精心编写的组装和制造操作而量身定制。如今，机器人智能体的应用正在扩展到生产线之外，并朝着协作活动、与其他机器人交互和协助人类的方向发展。这些代理能够探索非结构化环境，同时定位自己、绘制环境地图、行走并与人互动，但它们仍然缺乏处理简单对象所需的灵活性。不幸的是，使机器人代理在工厂生产线上取得成功的能力在家庭、学校和医院等非结构化环境中是不够的。健康和老年护理等应用或发生在危险或高风险环境中的应用要求机器人代理能够完成复杂而灵活的操作操作。为了执行这种活动，这些代理必须能够处理不同材料的各种物体，这些物体可以位于机器人的视野之外，或者被环境或机器人本身部分遮挡。识别和构建非遮挡对象的模型主要是使用图像处理、计算机视觉和机器学习方法完成的。另一方面，在手部操作过程中，杂乱环境中或部分遮挡的对象可能只由采用触觉感知策略的代理来识别、建模和灵活地操纵。从长远来看，该研究计划旨在开发新的触觉感知方法，用于异质对象的操作、识别、分类和建模，以及触觉信息的编码和交流。拟议的研究计划还考虑将其他感知模式，如计算机视觉、肌电信号和手势识别整合到协作和辅助机器人代理中。新的理论结果将在考虑到一系列应用的情况下产生，从虚拟和虚拟现实、人机界面和远程操纵到智能假肢。这一研究计划的结果将为卫生和工业部门的实际问题提供解决方案，并将有助于发展学术界和工业界之间的密切研究合作。在接下来的五年里，共有六名研究生和五名本科生将在这笔赠款的支持下接受培训，为加拿大的高素质人才做出贡献。该计划促成的研究成果将在主要的参考期刊上传播，并在该领域的国际会议上发表。

Traditional industrial robots are equipped with highly specialized sensing, perception, and actuation suites tailored to repeat a limited set of carefully scripted assembly and manufacturing operations. Nowadays, the application of robotic agents is expanding out of the production lines and advancing towards collaborating activities, interacting with other robots, and assisting humans. These agents have the ability to explore unstructured environments while simultaneously localizing themselves, mapping their surroundings, walking and interacting with people, but they still lack the dexterity required to handle even simple objects. Unfortunately, the capabilities that make robotic agents successful on factory lines are not sufficient in unstructured settings like homes, schools, and hospitals. Applications such as health and eldercare or that take place in hazardous or high-risk environments require robotic agents to be able to complete complex dexterous manipulation operations. To perform such activities, these agents must be able to handle various objects of different materials that can be situated out of the field of view of the robot or partially obstructed by the environment or by the robot itself. Identifying and constructing models for non-occluded objects is primarily accomplished using image processing, computer vision, and machine learning methods. On the other hand, objects in cluttered environments or partially occluded during in-hand manipulation may be identified, modeled and dexterously manipulated only by agents employing tactile perception strategies. In the long term, this research program aims at developing novel tactile perception methods for the manipulation, identification, classification, and modeling of heterogeneous objects, as well as the encoding and communication of tactile information. The proposed research program also contemplates the integration of other perceptual modalities such as computer vision, myoelectric signals, and gesture recognition into collaborative and assistive robotic agents. New theoretical results will be generated with a range of applications in mind, varying from virtual and virtualized reality, human-machine interfaces and telemanipulation to smart prostheses. The results of this research program will provide solutions for practical problems in the health and industrial sectors and will help develop close research collaborations between academia and industry. A total of six graduate and five undergraduate students will be trained with the support of this grant over the next five years, contributing to Canada's highly qualified personnel. The research findings enabled by the program will be disseminated in leading refereed journals and presented at international conferences of the field.