paper / richardssutton / May 12 / failedAverage-reward reinforcement learning in semi-Markov decision processes via relative value iteration