Human-Like Sensor Fusion Implemented in the Posture Control of a Bipedal Robot

Posture control represents the basis for many human sensorimotor activities such as standing, walking or reaching. It involves inputs from joint angle, joint torque, vestibular and visual sensors as well as fusions of the sensor data. Roboticists may draw inspirations from the human posture control methods when building devices that interact with humans such as prostheses or exoskeletons. This study describes multisensory fusion mechanisms that were derived from human perception of ego-motion. They were implemented in a posture control model that describes human balancing of biped stance during external disturbances. The fusions are used for estimating the disturbances and the estimates, in turn, command joint servo controls to compensate them (disturbance estimation and compensation, DEC, concept). An emergent property of the network of sensory estimators is an automatic adaptation to changes in disturbance type and magnitude and in sensor availability. Previously, the model described human and robot balancing about the ankle joints in the sagittal plane. Here, the approach is extended to include the hip joints. The extended human-derived model is again re-embodied in a biped posture control robot constructed with human anthropometrics. The robot is tested in direct comparison with human subjects. Results on hip and ankle sway responses to support surface rotation are described. Basic resemblance of the results suggests that the robot’s DEC controls capture important aspects of the human balancing.