Knowledge Representation (KR) is one of the prominent approaches to Artificial Intelligence (AI) that is concerned with representing knowledge in a form that computer systems can utilize to solve complex problems. Answer Set Programming (ASP), based on the stable model semantics, is a widely-used KR framework that facilitates elegant and efficient representations for many problem domains that require complex reasoning.

However, while ASP is effective on deterministic problem domains, it is not suitable for applications involving quantitative uncertainty, for example, those that require probabilistic reasoning. Furthermore, it is hard to utilize information that can be statistically induced from data with ASP problem modeling.

This dissertation presents the language LP^MLN, which is a probabilistic extension of the stable model semantics with the concept of weighted rules, inspired by Markov Logic. An LP^MLN program defines a probability distribution over "soft" stable models, which may not satisfy all rules, but the more rules with the bigger weights they satisfy, the bigger their probabilities. LP^MLN takes advantage of both ASP and Markov Logic in a single framework, allowing representation of problems that require both logical and probabilistic reasoning in an intuitive and elaboration tolerant way.

This dissertation establishes formal relations between LP^MLN and several other formalisms, discusses inference and weight learning algorithms under LP^MLN, and presents systems implementing the algorithms. LP^MLN systems can be used to compute other languages translatable into LP^MLN.

The advantage of LP^MLN for probabilistic reasoning is illustrated by a probabilistic extension of the action language BC+, called pBC+, defined as a high-level notation of LP^MLN for describing transition systems. Various probabilistic reasoning about transition systems, especially probabilistic diagnosis, can be modeled in pBC+ and computed using LP^MLN systems. pBC+ is further extended with the notion of utility, through a decision-theoretic extension of LP^MLN, and related with Markov Decision Process (MDP) in terms of policy optimization problems. pBC+ can be used to represent (PO)MDP in a succinct and elaboration tolerant way, which enables planning with (PO)MDP algorithms in action domains whose description requires rich KR constructs, such as recursive definitions and indirect effects of actions.

In this thesis, a new approach to learning-based planning is presented where critical regions of an environment with low probability measure are learned from a given set of motion plans. Critical regions are learned using convolutional neural networks (CNN) to improve sampling processes for motion planning (MP).

In addition to an identification network, a new sampling-based motion planner, Learn and Link, is introduced. This planner leverages critical regions to overcome the limitations of uniform sampling while still maintaining guarantees of correctness inherent to sampling-based algorithms. Learn and Link is evaluated against planners from the Open Motion Planning Library (OMPL) on an extensive suite of challenging navigation planning problems. This work shows that critical areas of an environment are learnable, and can be used by Learn and Link to solve MP problems with far less planning time than existing sampling-based planners.

While there are many existing systems which take natural language descriptions and use them to generate images or text, few systems exist to generate 3d renderings or environments based on natural language. Most of those systems are very limited in scope and require precise, predefined language to work, or large well tagged datasets for their models. In this project I attempt to apply concepts in NLP and procedural generation to a system which can generate a rough scene estimation of a natural language description in a 3d environment from a free use database of models. The primary objective of this system, rather than a completely accurate representation, is to generate a useful or interesting result. The use of such a system comes in assisting designers who utilize 3d scenes or environments for their work.

Classical planning is a field of Artificial Intelligence concerned with allowing autonomous agents to make reasonable decisions in complex environments. This work investigates
the application of deep learning and planning techniques, with the aim of constructing generalized plans capable of solving multiple problem instances. We construct a Deep Neural Network that, given an abstract problem state, predicts both (i) the best action to be taken from that state and (ii) the generalized “role” of the object being manipulated. The neural network was tested on two classical planning domains: the blocks world domain and the logistic domain. Results indicate that neural networks are capable of making such
predictions with high accuracy, indicating a promising new framework for approaching generalized planning problems.

Recent advancements in external memory based neural networks have shown promise

in solving tasks that require precise storage and retrieval of past information. Re-

searchers have applied these models to a wide range of tasks that have algorithmic

properties but have not applied these models to real-world robotic tasks. In this

thesis, we present memory-augmented neural networks that synthesize robot navigation policies which a) encode long-term temporal dependencies b) make decisions in

partially observed environments and c) quantify the uncertainty inherent in the task.

We extract information about the temporal structure of a task via imitation learning

from human demonstration and evaluate the performance of the models on control

policies for a robot navigation task. Experiments are performed in partially observed

environments in both simulation and the real world

Reinforcement learning (RL) is a powerful methodology for teaching autonomous agents complex behaviors and skills. A critical component in most RL algorithms is the reward function -- a mathematical function that provides numerical estimates for desirable and undesirable states. Typically, the reward function must be hand-designed by a human expert and, as a result, the scope of a robot's autonomy and ability to safely explore and learn in new and unforeseen environments is constrained by the specifics of the designed reward function. In this thesis, I design and implement a stateful collision anticipation model with powerful predictive capability based upon my research of sequential data modeling and modern recurrent neural networks. I also develop deep reinforcement learning methods whose rewards are generated by self-supervised training and intrinsic signals. The main objective is to work towards the development of resilient robots that can learn to anticipate and avoid damaging interactions by combining visual and proprioceptive cues from internal sensors. The introduced solutions are inspired by pain pathways in humans and animals, because such pathways are known to guide decision-making processes and promote self-preservation. A new "robot dodge ball' benchmark is introduced in order to test the validity of the developed algorithms in dynamic environments.