Reset-Free Reinforcement Learning via Multi-State Recovery and Failure Prevention for Autonomous Robots | TUP Journals & Magazine | IEEE Xplore