System and method for fault detection in robotic actuation

A system for fault detection in robotic actuation, the system comprising:

• a plurality of mobile robots 102, wherein each mobile robot from the plurality of mobile robots is associated with an actuation data gathering unit, wherein the actuation data gathering unit comprising a plurality of sensors;
• a plurality of computing devices 104, wherein each computing device comprising one or more memories comprising programmed instructions, a repository for storing the set of tasks associated with each mobile robot, wherein one or more hardware processors operatively coupled to the one or more memories, wherein the one or more hardware processors are capable of executing the programmed instructions stored in the one or more memories and a Deep Learning (DL) analytics unit, and wherein the DL analytics unit is configured to:
• receive, a set of tasks in the form of time series from the actuation data gathering unit associated with each mobile robot from the plurality of mobile robots, wherein the set of tasks comprises a set of actuation data associated with each mobile robot from the plurality of mobile robots;
• compute, a next activity associated with each mobile robot based on a DL analytics, wherein the DL analytics is based on an optimal resource allocation technique to minimize computation cost and learning time, by:
• (i) segmenting, each task from the set of tasks into a plurality of sub action windows, wherein an execution time associated with each sub action window is equal;
• (ii) selecting an optimal computing device from a plurality of computing devices based on an optimal load distribution technique, wherein the optimal load distribution technique is a master slave framework with a master computing device and a plurality of slave computing devices, and wherein the master slave framework is pre-trained to perform the optimal load distribution for executing a stateful Long Short Term Memory (LSTM);
• and
• (iii) computing, by the optimal computing device, the next activity associated with the plurality of robots by analyzing the plurality of sub action windows based on the pre-trained stateful LSTM, wherein computing the next activity associated with the plurality of robots comprises:
• receiving the plurality of sub action windows;
• computing a probabilistic trellis from the plurality of sub action windows, wherein the probabilistic trellis is computed by utilizing a set of prior probabilistic transitions recorded during robotic simulations;
• computing a set of posterior probability based on the probabilistic trellis, by performing a plurality of trails;
• and
• computing the next activity associated with the plurality of robots based on the set of posterior probability, using the pre-trained stateful LSTM;
• and
• detect, a fault associated with each mobile robot from the plurality of mobile robots by comparing, the next activity with the actual activity associated with the mobile robot.

One or more non-transitory machine readable information storage mediums for fault detection in robotic actuation, comprising:

• receive, by one or more hardware processors, a set of tasks in the form of time series from a plurality of robots, wherein the set of tasks comprises a set of actuation data associated with each robot from the plurality of robots;
• compute, by the one or more hardware processors, a next activity associated with each robot based on a Deep Learning (DL) analytics, wherein the DL analytics is based on an optimal resource allocation technique to minimize computation cost and learning time, by:
• (i) segment, each task from the set of tasks into a plurality of sub action windows, wherein an execution time associated with each sub action window is equal;
• (ii) select an optimal computing device from a plurality of computing devices based on an optimal load distribution technique, wherein the optimal load distribution technique is a master slave framework with a master computing device and a plurality of slave computing devices, and wherein the master slave framework is pre-trained to perform the optimal load distribution for executing a stateful Long Short Term Memory (LSTM);
• and
• (iii) compute, by the optimal computing device, the next activity associated with the plurality of robots by analyzing the plurality of sub action windows based on the pre-trained stateful LSTM, wherein computing the next activity associated with the plurality of robots comprises:
• receiving the plurality of sub action windows;
• computing a probabilistic trellis from the plurality of sub action windows, wherein the probabilistic trellis is computed by utilizing a set of prior probabilistic transitions recorded during robotic simulations;
• computing a set of posterior probability based on the probabilistic trellis, by performing a plurality of trails;
• and
• computing the next activity associated with the plurality of robots based on the set of posterior probability, using the pre-trained stateful LSTM;
• and
• detect, by the one or more hardware processors, a fault associated with each robot from the plurality of robots by comparing, the next activity with the actual activity of the robot.