An efficient dynamic straight-knee walking pattern generation method for bipedal robots on changing terrains is proposed. The method consists of two steps: path optimization and speed planning. In the path planning phase

a path encoded by low-dimensional parameters defined in the hybrid kinematic coordinate domain is used to describe the walking process. The present hybrid kinematic coordinate representation naturally ensures a straight-knee state for the supporting leg and prevents the swing leg from colliding with the ground in the single support phase. To obtain the locally dynamically feasible whole-body motion paths

the parameters of the torso's horizontal motion path with the objective of minimizing the single-step time is optimized

and the parameters of the swing foot path with the objective of maximizing the stability margin in transition frames is also optimized. After obtaining the path

an improved sparse grid forward-backward algorithm is used for speed planning to ensure that it satisfies the dynamic constraints. In a simplified bipedal system simulation experiment with 15 degrees of freedom

the present method can generate the dynamic straight-knee walking trajectories that can adapt to different terrain changes within 2.5 ms.