Multi-UAV Systems

该项目将研究支持人机团队的机器学习算法的开发。这些团队包括人类与软件代理或机器人一起工作，通过广泛的协作和竞争关系。在某些情况下，人类可能负责团队，而在其他情况下，代理人或机器人可能负责。正如我们之前的研究所显示的那样，这种形式的灵活自主要求机器具有主动性，并表现出学习能力，以便适应团队成员，特别是团队中人类成员的不同需求。设计这种算法所涉及的挑战包括：(i)处理稀疏数据集（ii）从有限的输入推断人类偏好（iii）随着时间和不同情况的学习。该项目将着眼于开发人机交互中使用的机器学习技术的调查，重点是强化学习技术和学习系统的人机界面设计。在此之后，我们将开发设计人机团队算法和学习算法的方法。算法的评估将在两种可能场景下的真实数据上进行：(i)辅助生活系统，人类依靠自主物联网平台监测患有痴呆症的家庭成员；（ii）多无人机系统，在非常动态的灾难响应环境中依赖最少一组飞行员。该项目符合该大学对人工智能和自主系统的关注。特别是，主管将与ECS和工程部的同事合作开发测试平台并评估设计的算法。该项目也与由主管指导的新创建的机器智能中心完全一致。

The project will investigate the development of machine learning algorithms that will support human machine teams. These teams involve humans working alongside software agents or robots through a broad range of collaborative and competitive relationships. At some points humans may be in charge of the team and at others, agents or robots may be in charge. As previously shown in our studies, such forms of flexible autonomy require machines to be proactive and exhibit learning abilities in order to adapt their responses to the varying requirements of the individual team members and particularly the human members of the team. The challenges involved in designing such algorithms include (i) dealing with sparse datasets (ii) infering human preferences from limited inputs and (iii) learning over time and in different situations. The project will look at developing a survey of machine learning techniques used in human-machine interaction, with a focus on re-inforcement learning techniques and human-machine interface design for learning systems. Following this, we will develop methodologies to design human-machine teaming algorithms and learning algorithms.The evaluation of the algorithms will be done on real data with two possible scenarios: (i) assisted living systems where humans rely on autonomous IoT platforms to monitor family members with dementia and (ii) multi-UAV systems that rely on a minimum set of pilots in very dynamic disaster response settings.This project fits within the university's focus on AI and Autonomous Systems. In particular, the supervisor will be collaborating with colleagues across ECS and Engineering to develop testbeds and evaluate the algorithms designed. The project also perfectly aligns with the newly created Centre for Machine Intelligence directed by the supervisor.