A novel foot slip detection algorithm using unscented Kalman Filter innovation

Abstract

A novel slip detection algorithm is proposed using the innovation term of the Unscented Kalman Filter (UKF). An intentional modeling error was introduced in the dynamic model of a block resting on a slope, including tilt angle and angular velocity. The model was formulated with an assumption of no translations in x-and y-directions. This model was implemented in the UKF based on gyro and accelerometer measurements. When the block slid, the UKF innovation increased considerably due to unmodeled dynamics (i.e., translation). The smoothed innovation was used to detect slip of the block, instead of using the metrics of estimation/measurement of the translational acceleration. As proof of concept, drag-sled stick-slip experiments were conducted under dry and wet surface conditions for level and inclined surfaces. Results indicate versatility of the proposed algorithm for slip detection using boosted innovation. Because accurate metrics of estimation/measurements were not required, parameter tuning was simple, and inexpensive MEMS-based sensors provided satisfactory data quality for slip detection without further error correction.

A novel method for slip prediction of walking biped robots

“…24 In the same contest, the acceleration and gyro readings with unscented Kalman filter UKF are used for slip detection, here the UKF innovation is used for detecting the slip. 25 For slow walking bipeds, the biped body acceleration is utilized through simplified models to calculate the applied forces at the foot. Then, the slip is detected by comparing the calculated applied forces with the measured forces from force sensors assembled at the foot sole.…”

State estimation for legged robots on unstable and slippery terrain

“…As mentioned earlier, the IMU measurements are linked to the rates of the filter states and are thus included into the continuous time differential equations of the prediction model. Using equation (15) and (16) and carefully evaluating the total derivatives we can write:…”

“…Most approaches use some additional force sensing on the foot level and compare desired and actual forces in order to detect slippage [11]. More recently, Okita and Sommer [16] considered slip events being anomalies which can be detected by employing appropriate filtering methods. In a simplified 2D stick-slip experiment they showed how to detect slippage using smoothed innovation in an Unscented Kalman filter (UKF) setup.…”