This study proposes a real-time sequential sensor fusion based gait pattern controller. During the walking cycle, the sensor data is analyzed and obtained by several inertial measurement units (IMUs) and pressure sensors on the robot. Besides, in order to model the correction values of the walking parameters, this approach applies long short-term memory (LSTM) to build a feedback system for a bipedal humanoid robot. For training the network, four sequential features including two-direction Center of Pressure (CoP) and two-direction acceleration are acquired by using pressure sensors and IMUs on the robot. While these features are mapped to the network inputs, the network output is mapped by the optimal solutions generated using particle swarm optimization (PSO). The proposed LSTM feedback network comprises four layers. The effect of the proposed methodology is first tested in a robot simulation environment and then tested on a real bipedal humanoid robot. The experimental results demonstrate that this methodology can make the humanoid robot perform better than when using a fixed gait. It also provides self-adjustment capability to the robot to enable it to avoid situations like falling.
All Science Journal Classification (ASJC) codes