Apollo is a robot built by Apptronik with a human-shaped torso, arms and legs. In a Google test, it learned to throw bean bags through a hole in a wooden board. Google first had its artificial intelligence watch a person demonstrate the throw.
The game is called cornhole: players stand some distance from a sloping board and try to toss a small, filled fabric bag through its hole. According to Google, the robot landed seven throws in a row during the test.
The throw is trickier than it looks. The arm needs to move at the right speed, and the fingers must open at just the right moment. Release the bag too early or too late, and it can miss the hole.
Practising on a computer first
To teach the skill, Google used its Gemini 4 Argon AI system. It watched the human demonstration, calculated the movements needed and built a simulated practice environment on a computer. The movements could be tried out there before the physical robot took its turn.
The successful computer-trained motion was then transferred to Apollo. The robot had to carry out the same movement with its real arms and hands.
Google aims to speed up the teaching of new robot skills with this approach. The company says setting up the practice environment for this test took a few hours, compared with weeks of manual work.
