Abstract: In order to realize the human-like motion control of the robotic joint system actuated by antagonistic pneumatic artificial muscles (PAMs), it is regarded as a non-linear optimal control problem with consideration of its serious non-linear properties, and the optimal performance indexes are determined according to the minimum jerk model based on the joint motion characteristics of the human arm. Firstly, the robotic joint model is linearized with the extended state observer. Then, an optimal control law is designed on the linearized standard integrator series model to realize the unconstrained human-like trajectory of the robotic joint. The simulation results show that the proposed algorithm can achieve human-like motion with a wide range (0~120°), the joint trajectories are insensitive to load variation (1 kg~5 kg), and it demonstrates good anti-disturbance performance in joint motion. The proposed approach is validated through physical experiments on an experimental platform of the robotic joint, and the design rules of the tuning parameters are discussed. Only two parameters are required to be tuned in the proposed algorithm. It is suitable for human-like motion control of the joints actuated by antagonistic PAMs, and can meet the requirements for the intrinsic safety, the motion compliance and the human-like motion patterns of collaborative robots.