Experience-based torque estimation for an industrial robot

Abstract

Robotic manipulation tasks often require the control of forces and torques exerted on external objects. This paper presents a machine learning approach for estimating forces when no force sensors are present on the robot platform. In the training phase, the robot executes the desired manipulation tasks under controlled conditions with systematically varied parameter sets. All internal sensor data, in the presented case from more than 100 sensors, as well as the force exerted by the robot are recorded. Using Transfer Entropy, a statistical model is learned that identifies the subset of sensors relevant for torque estimation in the given task. At runtime, the model is used to accurately estimate the torques exerted during manipulations of the demonstrated kind. The feasibility of the approach is shown in a setting where a robotic manipulator operates a torque wrench to fasten a screw nut. Torque estimates with an accuracy of well below ±1 N m are achieved. A strength of the presented model is that no prior knowledge of the robot's kinematics, mass distribution or sensor instrumentation is required.

Screw torque traceability control: industrial application

“…In any kind of production process, not only the task should be properly designed but also the measurement system should be properly implemented to keep a good product quality and an efficient flow. For Chen (2014) and Berger et al (2016), accurate measurement of external forces is also important in manipulation tasks involving tool-usage, the fastening torque and the angular displacement of the screw have to be closely monitored. Muelaner et al (2010) describe "the accuracy of the measurement system is lower than that of assembly design takes place", calling this statement as a phenomenon.…”

Parallelism Strategies for Big Data Delayed Transfer Entropy Evaluation

“…Additionally, Transfer Entropy (TE) exists, which allows identification of a cause-effect relationship by not accounting for straightforward and uniquely shared information [15]. TE has been applied to many complex problems from diverse research fields e.g., oscillation analysis [16], finance [17,18], sensors [19][20][21][22], biosignals [23,24], thermonuclear fusion [25], complex networks [26], geophysical phenomena [27,28], industrial energy consumption network [29] and algorithmic theory of information [30]. In addition, TE has been implemented in non-Gaussian distributions, such as: multivariate exponential, logistic, Pareto (type I-IV) and Burr distributions [31].…”

A model-based residual approach for human-robot collaboration during manual polishing operations

“…Safety is the inherent and most important feature of a robot that has to work close to or collaborate with human beings. Due to the limits of sensors 15 and robot motion capabilities, e.g., when the human moves faster than the robot can sense or counteract, undesired contacts or collisions with humans may occur. This is handled in the lowest (direct) layer of the control architecture, which implements sensorless collision detection, isolation, and reflex reaction based on our residual method [4,5,6], a model-based scheme that monitors the 20 generalized momentum of the robot.…”

“…A force/torque (F/T) sensor is used to decrease uncertainties due to friction, so as to improve force estimation when the robot is not in motion. Under similar operative conditions, learning-based approaches have also been applied for estimating exchanged external forces [14,15]. The estimated external force and the saturation of motor control torques can be used to keep the actual exchanged forces under a safety threshold [16].…”