Learning to Predict Humanoid Fall

Abstract

Falls are undesirable in humanoid robots, but also inevitable, especially as robots get deployed in physically interactive human environments. We consider the problem of fall prediction: to predict if the balance controller of a robot can prevent a fall from the robot's current state. A trigger from the fall predictor is used to switch the robot from a balance maintenance mode to a fall control mode. It is desirable for the fall predictor to signal imminent falls with sufficient lead time before the actual fall, while minimizing false alarms. Analytical techniques and intuitive rules fail to satisfy these competing objectives on a large robot that is subjected to strong disturbances and exhibits complex dynamics. We contribute a novel approach to engineer fall data such that existing supervised learning methods can be exploited to achieve reliable prediction. Our method provides parameters to control the tradeoff between the false positive rate and the lead time. Several combinations of parameters yield solutions that improve both the false positive rate and the lead time of hand-coded solutions. Learned solutions are decision lists with typical depths of 5–10, in a 16-dimensional feature space. Experiments are carried out in simulation on an ASIMO-like robot.

Skill Memory in Biped Locomotion

“…Deviations between the robot's state and the model's prediction indicate the presence of disturbances. Kalyanakrishnan et al [10] extends this work to the case in which robots are subjected to even larger disturbances. Karssen and Wisse [11] explore Principal Component Analysis to predict fall in the 6-DoF Meta robot.…”

“…However, his work also focuses the falling strategy problem. Both Kalyanakrishnan et al [10] and Hohn and Gerth [7] explored machine learning-based solutions. The former employed decision lists, while the latter explored Gaussian Mixture Models and Hidden Markov Models to learn predictors for classifying reflex in a simulated BARt robot.…”

“…Besides, analytical solutions may have difficulties in dealing with changes imposed by time such as the wear and tear of the robot hardware. Kalyanakrishnan et al [10] explores supervised machine learning to achieve fall predictor, specifically rule-based systems such as decision trees and lists. This type of systems is attractive because of their easy interpretability.…”

Fall Prediction for New Sequences of Motions

“…However, these methods do not scale well to humanoid robots with complex geometries and high degrees of freedom. Since it is difficult to model real world variables such as friction and wear and tear etc, a data-driven approach is used to collect sensor data of stable and unstable trajectories and classified to determine if a fall will occur [7], [8]. The prediction is done during the execution before a fall occurs.…”

Orienting in mid-air through configuration changes to achieve a rolling landing for reducing impact after a fall

“…Previous work has used machine learning techniques to predict falling [16], manual hand tuning to design fall sequences [17], as well as an abstract model to control a safe fall [18], [19], [20]. When a fall is intended, such as landing from a jump, human landing behavior is observed to be similar to a damped spring-mass system [21].…”