Robotics and Autonomous Systems (UoE Lead with HWU) (PhD) - 4 Years (Full-Time)

问题陈述：与工程和计算机科学领域相比，生物和化学领域在工具和任务自动化方面的装备严重不足。自动化任务扫描的能力有助于减少人为错误并提高实验速度。自动化还可以使优化框架应用于任务，这些任务通常可以重新构建我们看待流程的方式。已经进行了研究，以使这一过程自动化。这些目前的方法在创造人类和机器人之间的协作环境以实现实验室工作自动化方面的潜力不足。报告中介绍的工作依赖于商业上可用的机器人平台，这严重限制了合作的潜力；实验室环境往往是为人类设计的，这一限制加剧了这种限制。因此，这些机器人的设计被固定在了环境中。中提出的自动化流程为特定流程的快速自动化提供了希望。然而，化学和生物涉及一系列复杂的任务，这些任务在不同的实验中可能会略有不同。目前，化学和生物实验室中用于任务自动化的现成解决方案通常配备了专门针对特定任务的机械手。这些机械手不适合脆弱而困难的实验室工作。通常，这是因为生物和化学中使用的许多工具都是为与人类交互而设计的。灵巧操作是机器人学研究的一个领域，可以直接应用于化学和生物实验室工作中复杂的操作挑战。OpenAI[5]的最新研究已经被用于手操作策略的强化学习的灵活操作中。熟练的操作过程是在模拟环境中学习的，具有高度的随机性。然而，最近由提出的工作，展示了高水平的健壮性与无传感器，纯顺应的方法。最初在中建议的机械派生漏斗的使用被扩展为使用合成的运动基本体，以便可以通过顺序运动基本体执行操纵任务。另一个可以结合到解决实验室工作的灵巧操作任务中的领域是交互感知。基于前面所述的运动健壮性，交互式感知可以用来将实验室自动化机器人的健壮性扩展到不同的工具。实验室中使用的工具集是一个相对有限的组，然而，工具子集的变化使得纯粹的感知模型可能不足以实现无故障交互。实现无故障的实验室环境是值得注意的，也是非常有价值的，因为任何部署的系统都必须非常健壮，因为在许多情况下，糟糕的化学或生物处理将导致危险的后果。

Problem Statement: The fields of biology and chemistry are severely under-equipped compared to the fields of engineering and computer science for the tooling and automation of tasks. The ability to automate taskscan help reduce human error and increase the rate of experimentation. Automation can also enable optimization frameworks to be applied to tasks that can often re-frame how we look at processes. Research has been conducted to automate this process. These current approaches fall short in their potential for creating a collaborative environment between humans and robots for the automation of laboratory work. The work presented in rely on a commercially available robot platform that severely limits the potential for cooperation; this limitation is exacerbated by the lab environment which is often designed for human use. As a result, these robot designs are shoe-horned into the environment. Automation processes like that proposed in offer promise for rapid automation of a specific process. However, chemistry and biology involve a complex multitude of tasks that can vary slightly from experiment to experiment. Currently, off the shelf solutions for automation of tasks in chemistry and biology labs are often equipped with manipulators geared towards very specific tasks. These manipulators are ill-suited for the fragile and difficult topology of lab work. Typically, this is because a lot of the tooling used in biology and chemistry is designed for interaction with humans. Dexterous manipulation is an area of robotics research that can be directly applied to the intricate manipulation challenge of chemistry and biology lab work. State of the art research by OpenAI [5] has been conducted into dexterous manipulation using reinforcement learning for in hand manipulation policies. The process of dexterous manipulation is learnt with high levels of randomness in a simulation environment. However, recent work proposed by, demonstrates high levels of robustness with a sensor-less, purely compliant approach. The use of mechanically derived funnels, originally proposed in, is expanded for the use of motion primitives that are composed such that manipulation tasks can be executed through sequential motion primitives. Another area that can be incorporated into solving the dexterous manipulation task for laboratory work, is interactive perception. Grounded in the motion robustness stated previously, interactive perception can be utilized to expand the robustness of a lab automation robot to different tools. The set of tools used in a lab are of a relatively finite group, however, the variations in the sub-sets of tools are such that pure perception models may not be robust enough for failure-free interaction. Enabling a failure-free lab environment is to be noted and highly worthy, as any system deployed must be extremely robust given the numerous situations where poor chemical or bio handling would lead to dangerous consequences.