國立虎尾科技大學 |

Trustworthy Behavior Modeling and Decision Making for Autonomous Driving.

紀錄類型:	書目-語言資料,手稿 : Monograph/item
正題名/作者:	Trustworthy Behavior Modeling and Decision Making for Autonomous Driving./
作者:	Jiao, Ruochen.
面頁冊數:	1 online resource (165 pages)
附註:	Source: Dissertations Abstracts International, Volume: 85-11, Section: B.
Contained By:	Dissertations Abstracts International85-11B.
標題:	Computer engineering. -
電子資源:	click for full text (PQDT)
ISBN:	9798382763064

Trustworthy Behavior Modeling and Decision Making for Autonomous Driving.
Jiao, Ruochen.

Trustworthy Behavior Modeling and Decision Making for Autonomous Driving. - 1 online resource (165 pages)

Source: Dissertations Abstracts International, Volume: 85-11, Section: B.

Thesis (Ph.D.)--Northwestern University, 2024.

Includes bibliographical references

Recent advances in deep learning have significantly propelled the development of autonomous vehicles. However, these systems face critical challenges in system-level safety, robustness against uncertainty and perturbation, poor explainability of driving behaviors, and generalizability to corner cases. These challenges are common across various safety-critical and human-centered systems and stem from the extensive use of deep neural network (DNN)-based machine learning across functions such as perception, prediction, planning, and decision-making.This thesis focuses on creating trustworthy machine learning components and algorithms for cyber-physical systems, particularly autonomous driving system. It introduces methods aimed at enhancing explainability and robustness in behavior modeling and promoting safer decision-making through robust, knowledge-aware representations and system-level adaptations.Deep-learning-based components in autonomous driving systems are often associated with poor explainability and generalizability due to their nature of the black box and overfitting on limited training samples. We've incorporated domain-specific knowledge such as behavior patterns, physics, and natural language into deep learning models. We propose the first work to explicitly model and encode the driver's aggressiveness and other high-level behaviors into motion prediction modules by a semi-supervised latent space regularization. We further integrate the physics laws (kinematics) into the deep generative models for vehicle motion generation, greatly enhancing the controllable data augmentation for better generalizability in long-tailed scenarios. Our recent efforts extend to leveraging large language models for improving the explainability and generalizability of vehicle motion prediction and planning, marking a significant stride towards trustworthy autonomous driving solutions.We have dedicated efforts to enhancing safety and robustness throughout the autonomous driving pipeline, from behavior prediction to decision-making. We propose a pioneering work on adversarial training for robust trajectory prediction against malicious attacks, where rich contexts and the absence of class labels pose significant challenges. To this end, we propose a novel method that introduces semantic features into the adversarial training process by a semi-supervised adversarial autoencoder, enhancing the adversarial robustness and robust generalization to unseen scenarios. Besides training-time defense, we further design data-efficient and well-generalized run-time anomaly detection pipelines for vehicle motion prediction with semantics modeling and contrastive learning techniques, which effectively leverage the underlying prior knowledge and data distribution. In addition to robust representation learning for the safety of individual modules, we have been focusing on system-level adaptation and addressing the robustness of autonomous systems holistically. We propose an end-to-end (from perception to control) uncertainty-aware and adaptive framework against adversarial attacks on automatic lane-centering systems. We also address the safety and robustness issue of emerging technologies such as the application of large language models (LLMs) in decision-making for autonomous driving. We integrate in-context learning with a model predictive controller (MPC) verifier and state-machine-based interactive planning scheme to ensure the safety of LLM planners at both the control and behavior levels.This dissertation presents a comprehensive framework and methodologies that enhance the trustworthiness of machine learning across various components of the autonomous driving pipeline, aiming to address critical challenges through knowledge-aware behavior modeling, robust representation learning, and safe and adaptive system-level design.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798382763064Subjects--Topical Terms:

569006
Computer engineering.
Subjects--Index Terms:

Autonomous drivingIndex Terms--Genre/Form:

554714
Electronic books.

Trustworthy Behavior Modeling and Decision Making for Autonomous Driving.
LDR:05298ntm a22004217 4500 001 1145005
005 20240617111738.5
006 m o d
007 cr mn ---uuuuu
008 250605s2024 xx obm 000 0 eng d
020 $a 9798382763064
035 $a (MiAaPQ)AAI31242517
035 $a AAI31242517
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Jiao, Ruochen. $3 1470203
245 1 0 $a Trustworthy Behavior Modeling and Decision Making for Autonomous Driving.
264 0 $c 2024
300 $a 1 online resource (165 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 85-11, Section: B.
500 $a Advisor: Zhu, Qi.
502 $a Thesis (Ph.D.)--Northwestern University, 2024.
504 $a Includes bibliographical references
520 $a Recent advances in deep learning have significantly propelled the development of autonomous vehicles. However, these systems face critical challenges in system-level safety, robustness against uncertainty and perturbation, poor explainability of driving behaviors, and generalizability to corner cases. These challenges are common across various safety-critical and human-centered systems and stem from the extensive use of deep neural network (DNN)-based machine learning across functions such as perception, prediction, planning, and decision-making.This thesis focuses on creating trustworthy machine learning components and algorithms for cyber-physical systems, particularly autonomous driving system. It introduces methods aimed at enhancing explainability and robustness in behavior modeling and promoting safer decision-making through robust, knowledge-aware representations and system-level adaptations.Deep-learning-based components in autonomous driving systems are often associated with poor explainability and generalizability due to their nature of the black box and overfitting on limited training samples. We've incorporated domain-specific knowledge such as behavior patterns, physics, and natural language into deep learning models. We propose the first work to explicitly model and encode the driver's aggressiveness and other high-level behaviors into motion prediction modules by a semi-supervised latent space regularization. We further integrate the physics laws (kinematics) into the deep generative models for vehicle motion generation, greatly enhancing the controllable data augmentation for better generalizability in long-tailed scenarios. Our recent efforts extend to leveraging large language models for improving the explainability and generalizability of vehicle motion prediction and planning, marking a significant stride towards trustworthy autonomous driving solutions.We have dedicated efforts to enhancing safety and robustness throughout the autonomous driving pipeline, from behavior prediction to decision-making. We propose a pioneering work on adversarial training for robust trajectory prediction against malicious attacks, where rich contexts and the absence of class labels pose significant challenges. To this end, we propose a novel method that introduces semantic features into the adversarial training process by a semi-supervised adversarial autoencoder, enhancing the adversarial robustness and robust generalization to unseen scenarios. Besides training-time defense, we further design data-efficient and well-generalized run-time anomaly detection pipelines for vehicle motion prediction with semantics modeling and contrastive learning techniques, which effectively leverage the underlying prior knowledge and data distribution. In addition to robust representation learning for the safety of individual modules, we have been focusing on system-level adaptation and addressing the robustness of autonomous systems holistically. We propose an end-to-end (from perception to control) uncertainty-aware and adaptive framework against adversarial attacks on automatic lane-centering systems. We also address the safety and robustness issue of emerging technologies such as the application of large language models (LLMs) in decision-making for autonomous driving. We integrate in-context learning with a model predictive controller (MPC) verifier and state-machine-based interactive planning scheme to ensure the safety of LLM planners at both the control and behavior levels.This dissertation presents a comprehensive framework and methodologies that enhance the trustworthiness of machine learning across various components of the autonomous driving pipeline, aiming to address critical challenges through knowledge-aware behavior modeling, robust representation learning, and safe and adaptive system-level design.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Computer engineering. $3 569006
650 4 $a Engineering. $3 561152
650 4 $a Robotics. $3 561941
653 $a Autonomous driving
653 $a Deep learning
653 $a Machine learning
653 $a Motion prediction
653 $a Safety and robustness
653 $a Large language models
655 7 $a Electronic books. $2 local $3 554714
690 $a 0464
690 $a 0800
690 $a 0537
690 $a 0771
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a Northwestern University. $b Electrical and Computer Engineering. $3 1186892
773 0 $t Dissertations Abstracts International $g 85-11B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31242517 $z click for full text (PQDT)