ICML-98 Submission #26

Title: Learning to Drive a Bicycle using Reinforcement Learning and Shaping

Authors with addresses: Jette Randløv and Preben Alstrøm.
CATS, Niels Bohr Institute,
University of Copenhagen, Blegdamsvej 17,
DK-2100 Copenhagen, Denmark

    Abstract (250 word maximum): We here present and solve a real-world problem of learning to drive a bicycle. We solve the problem by online reinforcement learning using the Sarsa($\lambda$)-algorithm. Finally we solve the dual problem of learning to balance a bike and then drive to a goal. In our approach the reinforcement function is independent of what task the agent tries to learn to solve.
Keywords: Reinforcement Learning, real-world problem, shaping.

Email address of contact author: randlov@nbi.dk

Phone number of contact author: +45 353 25 273