CISE Research Instrumentation: High-Performance Vehicle with Visual and Deictic Control

9421532 Nelson该奖项旨在支持罗切斯特大学计算机科学和机械工程系的研究工作，以建立和控制一对小型越野车；一架由人驾驶，另一架由电脑控制。该设备的有趣之处在于低成本、越野能力和相对较高的速度（50公里/小时）。这两辆车将支持半自主控制、实时视觉处理、实时决策、控制学习和仿真的研究。研究问题集中在半自主或“指示”控制，其中视觉或符号命令传递给计算机控制的车辆进行局部解释。这种控制风格越来越受欢迎的困难，实时控制任务（如远程操作），其中既不是完全自主，也不是直接远程操作是可行的。罗切斯特将继续研究这个主题。指示控制并不能解决所有的控制问题：控制学习可以发挥重要作用。控制学习的两个方面将被研究：学习可调整的“运动技能”，以应对高速的情况，以及在更高的层次上，学习适当的动作序列。罗切斯特大学已经在研究这两种学习方式。目前的学习算法通常在模拟中接受初始训练，因此将使用已经进行的复杂模拟工作。＊＊＊

9421532 Nelson This award is to support the research effort for the University of Rochester's Computer Science and Mechanical Engineering Departments to build and control a pair of small, off-road vehicles; one human-piloted, and one computer-controlled. Intrinsically interesting aspects of the equipment are low cost, off-road capability, and relatively high speeds (50 kph). The two vehicles will support research on semi-autonomous control, real-time visual processing, real-time decision-making, control learning, and simulation. The research issues center around semi-autonomous or "deictic" control in which visual or symbolic commands are passed to the computer-controlled vehicle for local interpretation. This control style is increasingly popular for difficult, real-time control tasks (e.g. telemanipulation) in which neither full autonomy, nor direct teleoperation is feasible. Rochester will continue working on this topic. Deictic control does not solve all control problems: control learning can play an important role. Two aspects of control learning will be investigated: learning tunable "motor skills" necessary to cope with high-speed situations, and, at a higher level, learning sequences of appropriate actions. Both types of learning are already under investigation at Rochester. Today's learning algorithms often receive initial training in simulation, so sophisticated simulation work already underway will be used. ***