DeXtreme: Transfer of Agile In-hand Manipulation from Simulation to Reality

Recent work has demonstrated the ability of deep reinforcement learning (RL) algorithms to learn complex robotic behaviours in simulation, including in the domain of multi-fingered manipulation.