Learning Latent Dynamics for Planning from Pixels

12 Nov 2018Danijar HafnerTimothy LillicrapIan FischerRuben VillegasDavid HaHonglak LeeJames Davidson

Planning has been very successful for control tasks with known environment dynamics. To leverage planning in unknown environments, the agent needs to learn the dynamics from interactions with the world... (read more)

PDF Abstract

Evaluation results from the paper

Task Dataset Model Metric name Metric value Global rank Compare
Continuous Control DeepMind Cheetah Run (Images) PlaNet Return 650 # 1
Continuous Control DeepMind Cup Catch (Images) PlaNet Return 914 # 1
Continuous Control DeepMind Walker Walk (Images) PlaNet Return 890 # 1