System and method for automatic error recovery in robotic assembly

A method for controlling a robotic arm repeatedly performing an operation including an insertion of a component along an insertion line to assemble a product, wherein a configuration of the robotic arm includes a wrist having a motion with multiple degrees of freedom, wherein the method uses a processor coupled with stored instructions implementing the method, wherein the instructions, when executed by the processor carry out steps of the method, comprising:

• accepting measurements of a force sensor operatively connected to the wrist of the robotic arm, the measurements include data indicative of force experienced by the wrist of the robotic arm at different positions of the wrist of the robotic arm along the line of insertion;
• accepting a probabilistic relationship for the force experienced by the wrist of the robotic arm along the insertion line as a probabilistic function of the position of the wrist of the robotic arm along the line of insertion, wherein the probabilistic function is learned from measurements of the operation repeatedly performed by one or multiple robotic arms having the configuration of the robotic arm under the control;
• determining a current value of the force and a current value of the position along the line of insertion based on the measurements of the force sensor;
• determining the probability of the current value of the force conditioned on the current value of the position according to the probabilistic function;
• determining a result of anomaly detection based on the probability of the current value of the force;
• and
• controlling the robotic arm based on the result of anomaly detection, wherein the probabilistic function is a Gaussian process regression defined by values of mean and covariance of the force for each value of the position of the wrist of the robotic arm along insertion line, comprising:
• selecting the mean and the variance from the Gaussian process regression corresponding to the current value of the position;
• and
• determining the probability of the current value of the force according to a Gaussian distribution defined by the selected values of the mean and the variance.

A non-transitory computer readable storage medium embodied thereon a program executable by a processor for performing a method, the method comprising:

• accepting measurements of a force sensor operatively connected to a wrist of the robotic arm, the measurements include data indicative of force experienced by the wrist of the robotic arm at different positions of the wrist of the robotic arm along a line of insertion;
• accepting a probabilistic relationship for the force experienced by the wrist of the robotic arm along the insertion line as a probabilistic function of the position of the wrist of the robotic arm along the line of insertion, wherein the probabilistic function is learned from measurements of the operation repeatedly performed by one or multiple robotic arms having the configuration of the robotic arm under the control;
• determining a current value of the force and a current value of the position along the line of insertion based on the measurements of the force sensor;
• determining the probability of the current value of the force conditioned on the current value of the position according to the probabilistic function;
• determining a result of anomaly detection based on the probability of the current value of the force;
• and
• controlling the robotic arm based on the result of anomaly detection, wherein the probabilistic function is a set of local Gaussian process regressions, wherein each local Gaussian process regression corresponds to its cluster position along the insertion line and is defined by values of mean and covariance of the force for each value of the position of the wrist of the robotic arm along insertion line, comprising:
• selecting the mean and the variance from each of the local Gaussian process regression corresponding to the current value of the position to produce a set of local Gaussian distributions defined by selected pairs of the mean and the variance;
• determining a local probability of the current value of the force according to each of the local Gaussian distribution to produce a set of local probabilities;
• and
• determining the probability of the current value of the force conditioned on the current value of the position as a weighted combination of the local probabilities, wherein the weight for a local probability is a function of a distance from a cluster position of the corresponding local Gaussian process regression to the current value of the position.

A system for controlling a robotic arm performing an operation including an insertion of a component along an insertion line to assemble a product, wherein a configuration of the robotic arm includes a wrist having a motion with multiple degrees of freedom, comprising:

• an input interface configured to accept measurements of a force sensor operatively connected to the wrist of the robotic arm, the measurements include data indicative of force experienced by the wrist of the robotic arm at different positions of the wrist of the robotic arm along the insertion line;
• a memory configured to store a probabilistic relationship for the force experienced by the wrist of the robotic arm along the insertion line as a probabilistic function of the positions of the wrist of the robotic arm along the line of insertion, wherein the probabilistic function is learned from measurements of the operation repeatedly performed by one or multiple robotic arms having the configuration of the robotic arm under the control;
• and
• at least one processor configured to run executable components comprising an anomaly detector configured to
• determine a current value of the force and a current value of the position along the line of insertion based on the measurements of the force sensor;
• determine the probability of the current value of the force conditioned on the current value of the position according to the probabilistic function;
• and
• determine a result of anomaly detection based on the probability of the current value of the force;
• and
• a recovery controller configured to control the robotic arm based on the result of anomaly detection, wherein the probabilistic relationship defines a confidence interval for non-anomalous values of the force for each position of a sequence of positions on the insertion line, wherein each of the non-anomalous values of the force for the current position has a probability above a confidence level according to the probabilistic function learned for the current position, and wherein the anomaly detector declares the current value of the force anomalous when the current value of the force is outside of the confidence interval for non-anomalous values of the force of the current position.

A system for controlling a robotic arm performing an operation including an insertion of a component along an insertion line to assemble a product, wherein a configuration of the robotic arm includes a wrist having a motion with multiple degrees of freedom, comprising:

• an input interface configured to accept measurements of a force sensor operatively connected to the wrist of the robotic arm, the measurements include data indicative of force experienced by the wrist of the robotic arm at different positions of the wrist of the robotic arm along the insertion line;
• a memory configured to store a probabilistic relationship for the force experienced by the wrist of the robotic arm along the insertion line as a probabilistic function of the positions of the wrist of the robotic arm along the line of insertion, wherein the probabilistic function is learned from measurements of the operation repeatedly performed by one or multiple robotic arms having the configuration of the robotic arm under the control;
• and
• at least one processor configured to run executable components comprising an anomaly detector configured to
• determine a current value of the force and a current value of the position along the line of insertion based on the measurements of the force sensor;
• determine the probability of the current value of the force conditioned on the current value of the position according to the probabilistic function;
• and
• determine a result of anomaly detection based on the probability of the current value of the force;
• and
• a recovery controller configured to control the robotic arm based on the result of anomaly detection, wherein the input interface is configured to accept the probabilistic function and to store the probabilistic function in the memory, wherein the probabilistic function is learned by fitting a Gaussian process regression model to training data defined by the measurements of the operation repeatedly performed by one or multiple robotic arms having the configuration of the robotic arm under the control.