Artificial intelligence-actuated robot

A robot, comprising:

• a robot structural frame;
• a kinematic chain including a root joint connected to the structure frame, the kinematic chain comprising a plurality of joints and links arranged downstream from the root joint and at least one end effector;
• a plurality of actuators, wherein the actuators in the plurality of actuators produce actuator data indicating current actuator states and respond to actuator command data to drive the actuators;
• a plurality of tendons connected to a corresponding plurality of actuation points on the kinematic chain and to actuators in the plurality of actuators, the plurality of tendons arranged to translate actuator position and force to actuation points on the kinematic chain, wherein actuators in the plurality of actuators positioned on the structural frame or on the kinematic chain upstream from respective actuation points in the plurality of actuation points;
• a sensor configured to generate sensor data indicating in three dimensions and in time a position of the at least one end effector and of an object;
• and
• a controller in communication with the plurality of actuators and the sensor to operate the kinematic chain and having access to memory storing a system file identifying a set of neural network topologies, neural network execution circuits, configurable to execute neural network topologies in the set, and to apply feedback data derived from or including the actuator data and the sensor data as input to the neural network execution circuits, the controller including a program to load and unload neural network topologies from the set according to a specified a task to manipulate the object in response to the feedback data.

A method of operating a robot, the method comprising:

• reading actuator state data from a plurality of actuators having positions and forces that are translated to corresponding actuation points on a kinematic chain having a first branch connected to a first end effector;
• generating sensor data indicating positions of an object and the first end effector;
• generating feedback data from the actuator state data and the sensor data;
• inputting the feedback data to a trained neural network to generate actuation command data to operate the kinematic chain to perform a task, wherein the actuation command data compensates for losses in precision in performance of the task due to variability in translating positions and forces of the actuators to corresponding actuation points on the kinematic chain;
• and
• driving the plurality of actuators to operate the kinematic chain to perform the task based on the actuator command data.