자료유형  

 학위논문

Control Number  

0017160346

International Standard Book Number  

9798382610849

Dewey Decimal Classification Number  

629.8

Main Entry-Personal Name  

Wang, Sean J.

Publication, Distribution, etc. (Imprint  

[S.l.] : Carnegie Mellon University., 2024

Publication, Distribution, etc. (Imprint  

Ann Arbor : ProQuest Dissertations & Theses, 2024

Physical Description  

139 p.

General Note  

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

General Note  

Advisor: Johnson, Aaron M.

Dissertation Note  

Thesis (Ph.D.)--Carnegie Mellon University, 2024.

Summary, Etc.  

요약Unmanned ground vehicles offer great potential, but struggle in rough terrain environments due to their inability to reason about complex dynamics over longer horizons. Thus, driving strategies are often short-sighted and underutilize the robot's dynamic capabilities. This thesis addresses the challenges of long-horizon, dynamics-aware decision making in rough terrain, arising from the inability to model complex system dynamics. The findings reveal two key insights: 1) leveraging low-quality simulation data can reduce training requirements of real-world dynamics models; and 2) long-horizon decision making can still utilize imprecise dynamics models through careful handling of prediction uncertainty.We first discuss our model-based reinforcement learning approach for rough terrain navigation. This approach trains a dynamics model to predict the robot's trajectory over uneven terrain and capture prediction uncertainty. Decision making uses this model along with a closed-loop divergence constraint to aid in longer-horizon trajectory prediction and prevent exploitation of potentially problematic modeling errors. We show that this approach leads to robust, non-myopic driving strategies that take full advantage of the robot's capabilities.Next, we explore leveraging simulation to reduce real-world training data requirements. This culminates in an approach that uses a large variety of simulated systems to train a dynamics model that quickly and probabilistically adapts to any new, including real-world, target system using any available target system data. Using this model within an uncertainty-aware decision making framework results in safe, albeit low performance, driving upon initialization. As more target system observations are collected, the adaptive dynamics model becomes more tailored to the target system resulting in increased driving performance.Finally, we combine these concepts to form a non-myopic rough terrain navigation framework that can quickly and robustly adapt to new target systems. We show that upon initialization, this framework chooses conservative routes that avoids obstacles. However, after just one demonstration of driving over obstacles, the framework chooses more aggressive routes over obstacles that match the system's capabilities.

Subject Added Entry-Topical Term  

Robotics.

Subject Added Entry-Topical Term  

Information technology.

Index Term-Uncontrolled  

Model-based reinforcement learning

Index Term-Uncontrolled  

Rough terrain navigation

Index Term-Uncontrolled  

Sim2real

Index Term-Uncontrolled  

Terrain environments

Index Term-Uncontrolled  

Learning dynamics

Added Entry-Corporate Name  

Carnegie Mellon University Mechanical Engineering

Host Item Entry  

Dissertations Abstracts International. 85-11B.

Electronic Location and Access  

로그인을 한후 보실 수 있는 자료입니다.

Control Number  

joongbu:657492