Deriving Rewards for Reinforcement Learning from Symbolic Behaviour Descriptions of Bipedal Walking