仿肌肉反射的双足机器人直立抗扰策略研究

Bipedal robot push-recovery behavior counts on integrate control from central controller. In order to resist disturbance from the out-lab environment, model and control algorithm become more and more complicated. As a result, the burden of controller is increasing. But the push-recovery ability is not apparently increasing. Human push-recovery behavior does not only count on brain. The muscle reflex also gives contributions. Muscle reflex can control joint to resist small disturbance without brain. The project attempts to investigate the mechanism of human of small disturbance resisting strategy using muscle reflex, and implement it on bipedal robots, releasing the burden of controller. .The following three parts will be studied: 1) mechanical impedance of Human ankle joint, 2) the relationship between muscle reflex and disturbance, 3) construction of bipedal robot control model for push-recovery based on muscle reflex. .The two key problems for the project are: 1) how to tune ankle joint mechanical impedance using muscle reflex, and 2) how to implement human muscle reflex control mechanism on bipedal robots. .Main contributions of the project are: 1) quantitative analysis of ankle joint mechanical impedance tuning using muscle reflex, via EMG, and 2) simplification of push-recovery control algorithm using muscle reflex mechanism. .Consequently, the results of this project will show some possibilities of improvement for bipedal robot push-recovery ability and improve the development of powered prosthesis and exoskeleton.

双足机器人稳定直立靠统一控制的方式实现。为抵抗复杂扰动，模型被不断扩充，算法越亦复杂，导致控制器负担增大，而实际效果却不明显，研究陷入瓶颈。人体直立抗扰策略的形成不只是脑控制的结果，肌肉反射也发挥着重要作用。肌肉反射可独立于脑，控制关节自发抵抗扰动，降低脑的负担。本项目通过研究人体踝关节肌肉反射控制机制，提出一种直立抗扰策略，使机器人踝关节可自行抵抗微小外力扰动，减轻控制器负担。主要研究：人体踝关节机械阻抗特性、人体踝关节反射增益与扰动的关系和机器人直立抗扰模型构建。重点解决两个问题：肌肉反射调控踝关节机械阻抗的模型建立、人体反射控制机理到机器人系统的映射。创新点在于：通过表面肌电信号定量分析反射对踝关节阻抗参数的调节；采用肌肉反射机制使双足机器人踝关节自行抵抗扰动。本项目的结果可提高双足机器人抗扰性能，对动力假肢、外骨骼技术的发展也有一定帮助。

双足机器人靠统一控制方式抵抗复杂扰动，算法日趋复杂，控制器负担增大，而实际效果却不明显。人体直立抗扰并不是由大脑统一管理，底层肌肉反射控制可独立于脑，自行抵抗微小扰动。本项目主要研究人体踝关节肌肉反射控制机制，提出一种直立抗扰策略，使机器人踝关节可自行抵抗微小外力扰动，减轻控制器负担。前期理论研究发现：1）直立下，可使用踝关节角度变量代替肌梭传感信息建立肌肉反射模型；2）肌肉反射模型在扰动初期（前200ms）对肌肉激活的预测精度高，随着时间推移，模型预测精度逐渐下降；3）直立时，踝关节力矩可表述为胫骨前肌、腓肠肌和比目鱼肌肌肉激活的线性组合，联合肌肉激活模型，可获得形式上与阻抗控制相一致的神经力学模型。通过辨识踝关节阻抗参数，给出了基于肌肉激活的阻抗调控模型，并应用于踝关节动力假肢中，实现直立抗扰。将肌肉反射模型应用于外骨骼控制中，使用模糊插值推理调整模型参数，实现针对个体差异的自适应参数调整。后续研究将进一步考虑中长延时反射，提高模型适用性，可针对复杂扰动，预测人体肌肉激活。