Dynamically Dexterous Robots via Switched and Tuned Oscillators

这项研究涉及对组合简单振荡器的技术进行实证和理论探讨，以从现有的简单振荡器产生新的更复杂的机器人行为。之前在PI实验室资助的工作已经证明，简单的振荡器可以用来赋予机器人极高但狭隘的动态灵活性。本研究试图证明，这种现有行为机制的有效耦合技术可以赋予更广泛的行为范围。该经验项目针对的是一种名为“传感器驱动的动态拾取和放置”的机器人操作场景。在这种情况下，机器人使用扁平的桨来获得扔进工作空间的球；根据需要在杂乱的环境中平衡或击球；将它们放入够不着的目的地容器中；并在过程中抵抗意外的干扰。对振荡器的求助反映了对事件驱动操纵的几乎独家理论关注。这种对反馈控制的承诺解决了这样一个事实，即机器可靠性中的故障经常发生是因为一些事件，这些事件本质上不是不可恢复的，但严重违反了我们的模型，并且无法预测。对动态环境中错误检测和恢复的关注与十年前(仍在大力发展中的)精细运动规划文献和百年前(仍在大力发展中)动力系统文献中的某些概念和分析工具的关注相一致，这些文献已被证明在PI先前资助的工作中很有用。

The research involves empirical and theoretical inquiry into techniques for combining simple oscillators to produce new and more complex robot behaviors from existing simpler ones. Previously funded work in the PI's laboratory has already demonstrated that simple oscillators can be used to confer upon a robot superlative but narrow dynamical dexterity. The present study attempts to demonstrate that effective coupling techniques for such existing behavioral mechanisms can confer a much broader range of behavior. The empirical project addresses a robotic manipulation scenario termed ``Sensor Driven Dynamical Pick and Place.'' In this scenario, the robot uses a flat paddle to acquire balls thrown into the workspace; balance or bat them as required through a clutter; loft them into an out-of-reach destination receptacle; and resistunanticipated perturbations along the way. The recourse to oscillators reflects an almost exclusive theoretical focus on event driven manipulation. Such a commitment to feedback control addresses the fact that failures in machine reliability frequently occur because of events which are not intrinsically unrecoverable but which violate our models dramatically and cannot be anticipated. The focus on error detection and recovery in a dynamical setting brings into correspondence the concerns of the decade old (and still vigorously developing) fine motion planning literature and certain concepts and analytical tools from the century old (and still vigorously developing) dynamical systems literature that have proven useful in the PI's previously funded work.