Inclination estimation and balance of robot using Vestibular Dynamic Inclinometer

Traditionally, the sensor design for Inertial Sensing of Human movement/posture is done using an accelerometer-gyroscope combination while detection of static or dynamic activities is done using uniaxial accelerometers. A single axis accelerometer-gyroscope combination uses sensor fusion along with a Kalman Filter to estimate the inclination. The inclination is obtained by integrating angular velocity and acceleration combinations. Due to imperfect sensors, this leads to magnification of errors and drift over time. The sensor discussed in the paper, called the Vestibular Dynamic Inclinometer (VDI), uses two dual-axis accelerometers and one single axis gyroscope per axis to estimate the inclination. The concept of `equilibrium axis', the axis along which the robot is at equilibrium, is discussed. The inclination angle obtained from the VDI is relative to the equilibrium axis and thus is more desirable as a control input rather than the inclination angle relative to the absolute gravity vector (as obtained from the accelerometer-gyroscope combination). The inclination angle obtained is independent of acceleration experienced by the robot and is not obtained by integration of angular velocity or acceleration unlike traditional designs. The control strategies proposed in the paper are torque control and acceleration control. Torque control requires generation of torque at the point of contact (analogous to the hip and ankle generating balancing torque in humans) and acceleration/postural control requires the acceleration of point of contact (analogous to running in humans) to maintain the body at equilibrium. The paper also proposes Lyapunov based non-linear adaptive controllers for an inverted pendulum(which approximates a human body) for both the control strategies. The controllers guarantee asymptotic stability. Simulations are performed assuming noise in the sensors.