Control apparatus, robot, learning apparatus, robot system, and method

A robot comprising:

• a gripper configured to grip an object;
• an arm configured to move the gripper;
• a physically flexible portion provided in at least one of an intermediate position of the gripper at a joint between a gripping surface and the gripper, a position between the gripper and the arm, and an intermediate position of the arm, the physically flexible portion introducing an uncertainty in the position of the gripper and the object;
• and
• a sensor configured to detect a state of at least one of the flexible portion, a portion of the robot on a side where the object is gripped relative to the flexible portion, and a gripped object, wherein
• the robot receives an action to perform predetermined work on the object, based on state observation data comprising flexible related observation data comprising observation data regarding a state of at least one of the physically flexible portion, and a portion of the arm on a side of the physically flexible portion where the object is gripped relative to the physically flexible portion and including the gripped object, the action based on inputting the state observation data to a learning model learned in advance through machine learning comprising learning processing in which a state space of the robot and an action space of the robot are subjected to dimension reduction based on the uncertainty of the position of the gripper and the object introduced by the physically flexible portion.

A control method of controlling a robot, comprising:

• (i) obtaining state observation data comprising flexible related observation data regarding a state of at least one of a physically flexible portion, and a portion of the robot on a side of the physically flexible portion where an object is gripped relative to the physically flexible portion, and including the gripped object,
• wherein robot is provided with a gripper configured to grip the object and an arm configured to move the gripper, and comprises the physically flexible portion provided in at least one of an intermediate position of a gripper at a joint between a gripping surface and the gripper, a position between the gripper and the arm, and an intermediate position of the arm, the physically flexible portion introducing an uncertainty in a position of the gripper and the object;
• and
• (ii) controlling the robot so as to output an action to be performed by the robot to perform predetermined work on the object, in response to receiving the state observation data, based on output obtained from the state observation data obtained in operation (i), by using a learning model learned in advance through machine learning, the machine learning comprising learning processing in which a state space of the robot and an action space of the robot are subjected to dimension reduction based on the uncertainty of the position of the gripper and the object introduced by the physically flexible portion.