基于高斯过程回归的足式机器人触地检测研究

Research on Collision Detection of Foot Robot Based on Gaussian Process Regression

触地检测通常利用安装在足端的力传感器来实现,但这也增加了腿部的转动惯量和成本,且安装在足端的传感器在冲击作用下容易损坏。针对该问题,利用高斯过程回归预测的方法,建立单腿的逆动力学模型实现无力传感器的足端力估计和触地的检测;提出利用规划运动轨迹上的状态点而非整个工作空间上的状态点来学习动力学模型的策略;利用多个采样周期的角速度来学习机器人的逆动力学模型,避免了角加速度的计算。最后,通过Simulink对基于高斯过程回归预测模型的足端力检测进行了仿真,结果表明基于高斯过程回归的模型能有效地检测出机器人足端触地状态。

Touchdown detection is usually realized by force sensors mounted on the foot,but this also increases the moment of inertia and cost of the leg,and the sensor mounted on the foot is easy to be damaged under the impact.Aiming at this problem,a method of Gaussian process regression prediction is proposed to build a single-leg inverse dynamics model to estimate the foot force and detect the touchdown without sensor in this paper;A strategy of using state points on the planned trajectory rather than on the whole workspace to learn the dynamic model is proposed;The angular velocity of multiple sampling periods is used to learn the inverse dynamics model of the robot,which avoids the calculation of angular acceleration.Finally,the foot force detection model based on Gaussian process regression prediction model is simulated by Simulink,and the results show that the model based on Gaussian process regression can effectively detect the robot's foot contact state.