Learning and planning with latent space dynamics has been shown to be useful\nfor sample efficiency in model-based reinforcement learning (MBRL) for discrete\nand continuous control tasks. In particular, recent work, for discrete action\nspaces, demonstrated the effectiveness of latent-space planning via Monte-Carlo\nTree Search (MCTS) for bootstrapping MBRL during learning and at test time.\nHowever, the potential gains from latent-space tree search have not yet been\ndemonstrated for environments with continuous action spaces. In this work, we\npropose and explore an MBRL approach for continuous action spaces based on\ntree-based planning over learned latent dynamics. We show that it is possible\nto demonstrate the types of bootstrapping benefits as previously shown for\ndiscrete spaces. In particular, the approach achieves improved sample\nefficiency and performance on a majority of challenging continuous-control\nbenchmarks compared to the state-of-the-art.\n