Facebook AI Researchers Achieve a 107x Speedup for Training Virtual Agents

Facebook researchers developed a GPU-accelerated deep reinforcement learning model that achieves near 100 percent success in navigating a variety of virtual environments without a pre-provided map.