Robot introspection with Bayesian nonparametric vector autoregressive hidden Markov models

Abstract

Abstract-Robot introspection, as opposed to anomaly detection typical in process monitoring, helps a robot understand what it is doing at all times. A robot should be able to identify its actions not only when failure or novelty occurs, but also as it executes any number of sub-tasks. As robots continue their quest of functioning in unstructured environments, it is imperative they understand what is it that they are actually doing to render them more robust. This work investigates the modeling ability of Bayesian nonparametric techniques on Markov Switching Process to learn complex dynamics typical in robot contact tasks. We compare the Markov switching process, together with different observation model and the dimensionality of sensor data. The work was tested in a snap assembly task characterized by high elastic forces. The task consists of an insertion subtask with very complex dynamics. Our approach showed a stronger ability to generalize and was able to better model the subtask with complex dynamics in a computationally efficient way. The modeling technique is also used to learn a growing library of robot skills, one that when integrated with low-level control allows for robot online decision making. Supplemental info can be found at [1].

Fast, robust, and versatile event detection through HMM belief state gradient measures

“…For skill identification, our current approach ranks 2nd behind the tool breakage work that identified anomalies in structured milling processes. Our work did better than [18] and [5], albeit these works modeled more complex dynamical phenomena. Similar statements can be made about anomaly identification.…”

“…In [5], [7], HMM scoring L is used for skill identification. Given S trained models for S robot skills, scoring yields the log-likelihood of a sequence of observations at time t for a trained model s ∈ S. Scoring is defined as:…”

“…Motion skill encoding is based on the premise that similar skills yield similar sensory-motor signatures [7], [8], [23]- [25]. As such, an HMM model Π s is derived from training data for a robot skill s. Optimized models (those whose scores improve as a function of selection of covariance model for Gaussian observations and latent state complexity) elicit narrower distributions.…”

“…1) Pre-Recovery Performance: For pre-recovery performance computation, during each trial, we record the anomalies triggered by the testing metric and count its true positives, false positives and false negatives as illustrated in technique ID Accuracy AFF/DCC/CSM/SVM [26] 84.66% 1 sHDP-HMM [5] 89.50% RCBHT w/ multiclass SVM [18] 97.00% HMM w/GradientBased Measure [current] 98.40% Tool breakage SVM [13] 99.38% technique anomalyID Accuracy HMM,varying threshold [24] ∼ 80.00% MLP [8] 83.27% sHDP-VAR-HMM,mag metric [7] ∼ 85.00% sHDP-HMM [5] 87.50% RCBHT w/ multiclass SVM [18] 97.00% HMM, gradient metric (current) 100.00% technique reaction time sHDP-VAR-HMM,mag metric [7] 3.70% 2 HMM, gradient metric (current) 1.84%…”

Recovering from External Disturbances in Online Manipulation through State-Dependent Revertive Recovery Policies

“…If so, the identification is deemed correct, and the time required to achieve the correct classification recorded. At the end of the cross-validation period, a classification accuracy matrix is derived as well as the mean time threshold value (these results were reported in [12], in this paper we limit ourselves to report on the recovery robustness of the system).…”