We couldn’t pass up the opportunity to play ping pong with Omron’s FORPHEUS robot at ATX West. FORPHEUS was created by Omron and is the world’s 1st “robot table tennis tutor”. It’s able to predict the movements of the human player, the type of shot they plan to play, and when a smash is coming. FORPHEUS matches the player’s skill level and sends encouraging messages throughout the match.
The Technologies of FORPHEUS
FORPHEUS Motion Controller
The controller used in Forpheus is an OMRON NJ series which is typically used in production facilities. All user interface technologies such as the touch panels for starting and switching modes and the servo controllers for the motors also use OMRON’s latest technology. Decisions on timing and direction are made within a thousandth of a second and the directional error rate for the return balls is within ± 10 cm. Learn more about Omron’s NJ Machine Controller Series
Machine Learning by AI
Through machine learning FORPHEUS can improve its ability to judge a players ability. By evaluating the player’s movements and the motion of the ball, the AI can gauge whether the opponent is experienced or a beginner. Forpheus then adjusts its ability level to ensure a successful rally.
Image Detection Using 3 Cameras
FORPHEUS has a three-camera system to identify the ball and evaluate players. The cameras mounted on the left and right allow the robot to see the ping pong balls in 3D, identical to the human visual system. FORPHEUS’ 3rd camera is in the center and is responsible for judging the players ability from their movements.
FORPHEUS Arm Control
Forpheus’ high speed robotic arm moves in response to its AI controller and was re-purposed from a robot arm used in packaging. The arm is controlled by the AI through a 5-axis motor system to swing the paddle at the ball. Paddle grip, ball hit location and arm position are all controlled to within 0.1mm.
FORPHEUS Predicts Where The Ball Will Go
Forpheus can predict the trajectory of the ball the moment a player hits it by calculating the speed and rotation of the ball up to 80 times per second. The robot moves its paddle to a position where the ball will be when it needs to return it and then calculates the trajectory of the ball it will be hitting back.