I'll believe it when I see a video of the “final version”.

Until then it's just like any other flaky and slow robot grasper demos out there trained using various flaky deep reinforcement learning algorithms.