Robot-object contact perception using symbolic temporal pattern learning

Abstract

Abstract-This paper investigates application of machine learning to the problem of contact perception between a robot's gripper and an object. The input data comprises a multidimensional time-series produced by a force/torque sensor at the robot's wrist, the robot's proprioceptive information, namely, the position of the end-effector, as well as the robot's control command. These data are used to train a hidden Markov model (HMM) classifier. The output of the classifier is a prediction of the contact state, which includes no contact, a contact aligned with the central axis of the valve, and an edge contact. To distinguish between contact states, the robot performs exploratory behaviors that produce distinct patterns in the time-series data. The patterns are discovered by first analyzing the data using a probabilistic clustering algorithm that transforms the multidimensional data into a onedimensional sequence of symbols. The symbols produced by the clustering algorithm are used to train the HMM classifier. We examined two exploratory behaviors: a rotation around the xaxis, and a rotation around the y-axis of the gripper. We show that using these two exploratory behaviors we can successfully predict a contact state with an accuracy of 88 ± 5 % and 81 ± 10 %, respectively.

Human skill capture: A Hidden Markov Model of force and torque data in peg-in-a-hole assembly process

“…HMM has also been used for state recognition in other applications. For instance, in [15], they proposed a method to recognise the states in valve opening. They used a symbolic representation of the F/T signal where the number of data is hugely reduced to a pre-defined number of segments.…”

Symbolic-Based Recognition of Contact States for Learning Assembly Skills

“…The results generated were compared with the most relevant work from the literature. In this regard, the method introduced by Jamali et al ( 2014 ) achieved an overall accuracy of 81%, and 85% for rotation about the x-axis and the y-axis, respectively. The HMM-PAA models proposed in this paper has an accuracy of 94% and is, therefore, an improvement.…”

“…This observation implies that there is an inverse relationship between the clearance and the accuracy of the CS recognition. This is due to the higher physical constraints Proposed approach (PAA) Proposed approach (K-means) MML-GMM (Jamali et al, 2014) EM-GMM (Jasim et al, 2017) HMM (Hannaford and Lee, 1990) PCA ( in a tight clearance insertion process, providing a betterdefined boundary that separates the consecutive CSs. The model trained based on tight clearances peg is more robust against geometrical variation.…”

“…The drawback of this work was the simplicity of the part's geometries used to validate the proposed approach. Jamali et al (2014) presented a CS learning algorithm based on a symbolic representation of temporal behavior during robot valve opening process where force signals were clustered using the Minimum Message Length (MML) (Wallace and Dowe, 2000). The labeled symbolic data were used to train an HMM to recognize the CS.…”

“…Confusion matrix of HMM trained with PAA 30 segments.the proposed approach was O(M), that only depends on M, while the complexity of MML-GMM depends on M multiplied by N samples and D. The overall performance appears similar in both methods. However, the method proposed inJamali et al (2014) requires additional exploration stage (set of random movements on x and y direction) before starting the recognition stage. Henceforth, it might require a longer time until it converges.…”