Intermittent predictive control of man and machine

五十年来，伺服机构作为一种简单的、反应性的、连续反馈的系统，一直被用作人类神经控制系统的模型。PID伺服是最著名的例子，它使用位置和速度反馈来稳定重要变量。我们一直都知道，神经系统比这复杂得多。我们都意识到我们的预期：有时我们的预期中的错误会让我们暴露出来，比如我们用了太多的力去提一个比我们预期的轻的手提箱。越来越多的人接受神经系统预测我们的世界，并预测神经信号如何转化为身体运动。作为电机控制的一种模式，伺服模式已经越来越多地被最优控制或连续预测控制模式所取代，这是建立在内部模型、预测和优化的工程控制方法之上的。虽然它在表达上更加复杂和丰富，但连续预测控制范式仍然与生物行为的几个方面不一致。生物防治本质上是可变的。对于人类来说，对刺激的时间反应是不一致的，对近期目标的重新表述是高度灵活的，有意义控制的带宽（即频率限制）是相当低的。这些神经特征不是源于工程洞察力的连续控制范式的自然结果，这种控制范式被精确地设计为否定人类控制的这些局限性，并产生具有高带宽、高度指定功能和最小目标灵活性的暂时一致的控制。这一建议的力量来自于一种新的工程控制方法，即间歇性预测控制。间歇预测控制在连续时间和离散时间控制的两个极端之间提供了一系列可能性：控制信号由一系列（连续时间）参数化轨迹组成，其参数间歇性地调整。它与离散时间控制的不同之处在于，样本之间的控制不是恒定的；它与连续时间控制的不同之处在于，轨迹是间歇性重置的。作为一类控制理论，间歇预测控制比连续控制具有更广泛的通用性，它提供了一种将连续预测和最优控制与间歇开环（弹道）控制相结合的新范式。这种新的间歇预测控制模式具有重要的技术应用价值。因此，有必要发展这些控制器的概念、理论和系统识别。这种新模式也直观地类似于人类生理控制系统，因为低带宽，灵活，可变控制是该机制的自然产物。我们打算发现人类的姿势机制是否最好地解释为连续PID类型的控制器或间歇控制过程。当姿态受到干扰时，发生的强大的纠正反应可以在高带宽连续反馈的基础上解释。然而，我们质疑这种机制是否在对未受干扰的、熟练的和习得的姿势活动（如站立）的精细控制中占主导地位。如果间歇性预测控制范式是适用的，那么自然的姿势平衡就被正确地重新解释为像任何其他形式的运动一样的中央调节、自愿控制。澄清这一问题将对多种医疗保健主题具有重要意义，包括无法站立的脊髓损伤患者的康复以及有跌倒史的老年患者的危险因素诊断。我们的目标是将我们的生物学见解融入到模拟人类神经系统实时灵活性的工程控制器的设计中。

For fifty years, the servo mechanism, a simple, reactive, continuous feedback system, has been used as a model of human neural control systems. The PID servo is the most well known example which uses positional and velocity feedback to stabilise the variable of importance. It has always been known that the nervous system is more sophisticated than this. We are all aware that we anticipate: sometimes the error in our expectations catches us out, such as when we use too much force to lift a suitcase that is lighter than we expected. There has been increasing acceptance that the nervous system predicts our world and predicts how neural signals will be converted into bodily movement. As a model of motor control, the servo paradigm has been increasingly replaced by the optimal control or continuous predictive control paradigm which is founded on the engineering control methodology of internal models, prediction and optimisation. While it is more sophisticated and richer in its expression, the continuous predictive control paradigm is still inconsistent with several aspects of biological behaviour. Biological control is inherently variable. For humans, the temporal response to stimuli is inconsistent, the reformulating of immediate goals is highly flexible, and the bandwidth, i.e. frequency limit, of meaningful control is rather low. These neural features are not natural outcomes of the continuous control paradigm derived from engineering insight, which has been designed precisely to negate these limitations of human control and produce control which is temporally consistent, with high bandwidth, highly specified function and hence minimal goal flexibility. This proposal's power derives from a new type of engineering control methodology known as Intermittent predictive control . Intermittent predictive control provides a spectrum of possibilities between the two extremes of continuous-time and discrete-time control: the control signal consists of a sequence of (continuous-time) parameterised trajectories whose parameters are adjusted intermittently. It is different from discrete-time control in that the control is not constant between samples; it is different from continuous-time control in that the trajectories are reset intermittently. As a class of control theory, intermittent predictive control is more general than continuous control and provides a new paradigm incorporating continuous predictive and optimal control with intermittent, open loop (ballistic) control. This new intermittent predictive control paradigm has important technological applications. Consequently there is a need to develop the concept, theory and system identification of these controllers. This new paradigm is also intuitively similar to human physiological control systems in that low bandwidth, flexible, variable control is a natural product of the mechanism. We intend to discover whether human postural mechanisms are best explained by a continuous PID type of controller or an intermittent control process. The powerful corrective responses that occur when posture is perturbed can be explained on the basis of high bandwidth continuous feedback. However, we question whether such mechanisms dominate in the exquisitely fine control of unperturbed, skilled and learned postural activities such as standing. If the intermittent predictive control paradigm is applicable, then natural postural balance is correctly reinterpreted as centrally modulated, voluntary control like any other form of movement. Clarification of this issue will have important implications for diverse healthcare topics including the rehabilitation of spinally injured patients who are no longer able to stand and the diagnosis of risk factors in elderly patients with a history of falling.We aim to incorporate our biological insights into the design of engineering controllers that mimic the real-time flexibility of the human nervous system.