A Latent State-Based Multimodal Execution Monitor with Anomaly Detection and Classification for Robot Introspection

Abstract

Robot introspection is expected to greatly aid longer-term autonomy of autonomous manipulation systems. By equipping robots with abilities that allow them to assess the quality of their sensory data, robots can detect and classify anomalies and recover appropriately from common anomalies. This work builds on our previous Sense-Plan-Act-Introspect-Recover (SPAIR) system. We introduce an improved anomaly detector that exploits latent states to monitor anomaly occurrence when robots collaborate with humans in shared workspaces, but also present a multiclass classifier that is activated with anomaly detection. Both implementations are derived from Bayesian non-parametric methods with strong modeling capabilities for learning and inference of multivariate time series with complex and uncertain behavior patterns. In particular, we explore the use of a hierarchical Dirichlet stochastic process prior to learning a Hidden Markov Model (HMM) with a switching vector auto-regressive observation model (sHDP-VAR-HMM). The detector uses a dynamic log-likelihood threshold that varies by latent state for anomaly detection and the anomaly classifier is implemented by calculating the cumulative log-likelihood of testing observation based on trained models. The purpose of our work is to equip the robot with anomaly detection and anomaly classification for the full set of skills associated with a given manipulation task. We consider a human-robot cooperation task to verify our work and measure the robustness and accuracy of each skill. Our improved detector succeeded in detecting 136 common anomalies and 368 nominal executions with a total accuracy of 91.0%. An overall anomaly classification accuracy of 97.1% is derived by performing the anomaly classification on an anomaly dataset that consists of 7 kinds of detected anomalies from a total of 136 anomalies samples.

Multi-Agent Systems for Search and Rescue Applications

“…Such a strategy could allow human operators to aid in difficult navigation or identification tasks for multiple robots without incurring the full cognitive load burden of direct teleoperation. Leveraging learned models of human preferences (i.e., when in a given scenario they would most likely take over direct teleoperation of an agent) [67,68] and learned models of when robots are likely to require intervention (e.g., by assessing incoming sensory data for statistical anomalies [69] or by incorporating human control expected utility into a learned policy [70]) in order to automate the sliding autonomy control handoff could further reduce operator training requirements and attention burden.…”

Endowing Robots with Longer-term Autonomy by Recovering from External Disturbances in Manipulation Through Grounded Anomaly Classification and Recovery Policies

“…The knowledge-based models, diminished the need for data and labeling. In [14], the sticky version of HDP was used along a vector autoregressive observation model to do anomaly identification and multi-class classification with a dynamic threshold that updates its parameters according to the latent state of a sub-task. In [15], anomaly identification was conducted with HMMs whose detection threshold varied according to clusters of execution progress.…”

Incremental Learning Introspective Movement Primitives From Multimodal Unstructured Demonstrations

“…Obviously, when these four parameters are known, the method is computationally feasible. The detailed options is assigned in Section IV through experiments and more detailed derivation process of the formulation can be found in our previous work [19]. So for, we can model the underlying dynamics from a set of unstructured demonstrations such that the problem turns into how to learn the unknown parameters in approximate fashion?…”

“…, K } denotes how many times of movement B i is successfully executed. Therefore, the probability mass function of transition p(T B i ,B * ) is formulated as a multinomial distribution that model a total N = i=K i=1 N i reverse executions, that is, (19) where, θ i indicates the probability of movement primitive B i is selected, which subject to θ i ∈ [0, 1] and K i=1 θ i = 1. Therefore, we use the multinomial distribution not only to intuitively depict the expectation of human intention on the recovery behavior when an abnormal occurrence is detected in human robot interaction scenarios, but also to provide an indirect way to express the intuitive understanding of human for expecting the motion of the robot end-effector, the related manipulation objects as well as the complex relationship between ''human-robot-environment''.…”

“…To address the problem of the representation of robot complex, multi-step task, we conventionally segment the task into multiple sub-tasks or movements manually and formulate them in a directed graph fashion, which lead to the weak exploration capability. This paper address how the IMPs can be acquired by incrementally learning from unstructured demonstrations, which is based on our previous Sense-Plan-Act-Introspect-Recover (SPAIR) system, details in [19]. The SPAI is fist designed for multimodal perception and learning during robot manipulation task, which easily record and synchronize the multimodal signals from different sensors in varying frequency.…”