Josh Merel

Both when studying animals and when constructing artificial agents, I am interested in computations underlying sensation and perception of the environment, selection and planning of behavior, and control of bodies to execute movement. At DeepMind, much of my research aims to build agents that receive ecologically plausible sensory inputs and coordinate relatively high-dimensional bodies (simulated using a physics engine) to solve tasks by producing complex, diverse, and adaptive behavior. See this relatively recent perspective article for more detail about how I think about and approach these problems.