In the emerging domain of self-aware and autonomous systems, the causal representation of the variables and the learning of their dynamics to make inferences of the future state at multiple abstraction levels in successive temporal slices, are getting attention of the researchers. This work presents a novel data-driven approach to learn the causal representation of the dynamic probabilistic graphical model. The proposed method employed the geometric-based approach to define the set of clusters of similar Generalized State (GS) space as linear attractors. Clustering of data variables corresponding to the linear attractors defines a set of switching vocabulary, which provides the higher-level representation of the graphical model, i.e., discrete and continuous levels. The transitions between the switching vocabulary are represented with the transition matrix, estimated from the temporal data series in switching models based on GSs. Each learned representation of the dynamic probabilistic graphical model is stored in Autobiographical Memory (AM) layers. A Markov Jump Particle Filter (MJPF) is proposed to make inferences at multiple abstraction levels of graphs which facilitates the detection of anomalies. Anomalies indicate that the agent encounters new experiences which can be learned incrementally and evolve new layers of AM. The proposed approach is extended for the learning of interactions between autonomous agents to make them self-aware. In Low dimensional case, data from the odometry trajectories and the control parameters, i.e., steering angle and rotors' velocity, is employed. However, data from the LiDAR, i.e., 3D point clouds, is used for the high-dimensional case. The deep learning approach, such as 3D Convolutional Encode-Decoder together with the transfer learning employed to extract the features from the LiDAR’s point clouds. A similar learning approach (mentioned above) is employed to detect anomalous situations. Three predictive models, i.e., piecewise nonlinear, piecewise linear, and nonlinear models, are proposed to analyse the multiple abstraction level anomalies, i.e., continuous level, discrete level, and voxel level. Concurrently, the public KITTI dataset from the complex/urban environment is employed to validate the proposed methodology. Qualitative and quantitative analysis of the proposed methodology is perform by employing the anomaly measurements and the ROC curves to estimate the accuracy, respectively.

Learning of Geometric-based Probabilistic Self-Awareness Model for Autonomous Agents

IQBAL, HAFSA
2022-05-26

Abstract

In the emerging domain of self-aware and autonomous systems, the causal representation of the variables and the learning of their dynamics to make inferences of the future state at multiple abstraction levels in successive temporal slices, are getting attention of the researchers. This work presents a novel data-driven approach to learn the causal representation of the dynamic probabilistic graphical model. The proposed method employed the geometric-based approach to define the set of clusters of similar Generalized State (GS) space as linear attractors. Clustering of data variables corresponding to the linear attractors defines a set of switching vocabulary, which provides the higher-level representation of the graphical model, i.e., discrete and continuous levels. The transitions between the switching vocabulary are represented with the transition matrix, estimated from the temporal data series in switching models based on GSs. Each learned representation of the dynamic probabilistic graphical model is stored in Autobiographical Memory (AM) layers. A Markov Jump Particle Filter (MJPF) is proposed to make inferences at multiple abstraction levels of graphs which facilitates the detection of anomalies. Anomalies indicate that the agent encounters new experiences which can be learned incrementally and evolve new layers of AM. The proposed approach is extended for the learning of interactions between autonomous agents to make them self-aware. In Low dimensional case, data from the odometry trajectories and the control parameters, i.e., steering angle and rotors' velocity, is employed. However, data from the LiDAR, i.e., 3D point clouds, is used for the high-dimensional case. The deep learning approach, such as 3D Convolutional Encode-Decoder together with the transfer learning employed to extract the features from the LiDAR’s point clouds. A similar learning approach (mentioned above) is employed to detect anomalous situations. Three predictive models, i.e., piecewise nonlinear, piecewise linear, and nonlinear models, are proposed to analyse the multiple abstraction level anomalies, i.e., continuous level, discrete level, and voxel level. Concurrently, the public KITTI dataset from the complex/urban environment is employed to validate the proposed methodology. Qualitative and quantitative analysis of the proposed methodology is perform by employing the anomaly measurements and the ROC curves to estimate the accuracy, respectively.
File in questo prodotto:
File Dimensione Formato  
phdunige_4625216.pdf

embargo fino al 27/05/2023

Descrizione: Main article
Tipologia: Tesi di dottorato
Dimensione 18.93 MB
Formato Adobe PDF
18.93 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11567/1081940
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact