Bayesian Inference Based Generation of Human-Like Anthropomorphic Arm Movements

Abstract

The addition of human-like gestures to robots can improve productivity and have a big influence on how people interact with them. In order to help anthropomorphic arms produce precise human-like motions, this research introduces a unique planning method for human-like movements. Movement primitives, a Bayesian network (BN), and a new Long Short Term Memory network make up the three elements of the approach (LSTM). Human arm movements are broken down using movement primitives, and classifying arm motions improves the realism of human-like motions. The probability that a movement will occur is predicted using a motion-decision algorithm based on Bayesian Netwrok, which also chooses the best mode of motion. A brand new LSTM is also created to address the inverse kinematics issues with anthropomorphic arms. Several models are combined into a single network using the LSTM, which modifies the network’s structure to reflect the properties of the individual models. The suggested method enables anthropomorphic arms to accurately produce a variety of human-like actions. Lastly, simulations are used to verify the effectiveness of the suggested technique.