Human-like Sensor Fusion Mechanisms in a Postural Control Robot

In humans, maintaining body posture is a basis for many activities such as standing, walking or reaching. Human posture control involves multi-sensory integration mainly of joint angle, joint torque, vestibular and visual inputs. This integration provides humans with high flexibility and with robustness in terms of failsafety. Roboticists may draw inspirations from the human control methods when building devices that interact with humans, such as prostheses or exoskeletons. This study presents a multisensory control method derived from human experiments, which is re-embodied in a biped postural control robot. The robot uses ankle and hip joints for balancing in the sagittal plane during external disturbances such as support surface motion. For the balancing, the robot estimates the external disturbances that have impact on its body by fusing the sensory signals. It then uses these estimates in negative feedback to command the local joint controls to compensate for the disturbances. This study describes the human sensor fusion mechanisms and their implementation into the robot, and it compares robot and human responses to support surface tilt. Measured balancing responses of the robot resemble in the main characteristics those of the human subjects, suggesting that the described sensor fusion mechanisms capture important aspects of human balancing.