Provided are a speed planning method and apparatus for self-driving, a device and a medium, relating to artificial intelligence technologies such as self-driving and deep learning. The solution includes acquiring the current state of a vehicle and interpolating actions based on the current state and a predetermined state table to obtain the target action of the current state. The current state includes at least the remaining running distance and the current speed. The state table is determined based on a reinforcement learning method and includes multiple states and an action performed in each state. The action performed in each state includes at least acceleration.