RT Dissertation/Thesis
T1 Learning Multi-Modal Self-Awareness Models Empowered by Active Inference for Autonomous Vehicles
A1 Nozari, Sheida
AB For autonomous agents to coexist with the real world, it is essential to anticipate the dynamicsand interactions in their surroundings. Autonomous agents can use models of the humanbrain to learn about responding to the actions of other participants in the environmentand proactively coordinates with the dynamics. Modeling brain learning procedures ischallenging for multiple reasons, such as stochasticity, multi-modality, and unobservantintents. A neglected problem has long been understanding and processing environmentalperception data from the multisensorial information referring to the cognitive psychologylevel of the human brain process. The key to solving this problem is to construct a computingmodel with selective attention and self-learning ability for autonomous driving, which issupposed to possess the mechanism of memorizing, inferring, and experiential updating,enabling it to cope with the changes in an external world. Therefore, a practical selfdrivingapproach should be open to more than just the traditional computing structure ofperception, planning, decision-making, and control. It is necessary to explore a probabilisticframework that goes along with human brain attention, reasoning, learning, and decisionmakingmechanism concerning interactive behavior and build an intelligent system inspiredby biological intelligence.This thesis presents a multi-modal self-awareness module for autonomous driving systems.The techniques proposed in this research are evaluated on their ability to model proper drivingbehavior in dynamic environments, which is vital in autonomous driving for both actionplanning and safe navigation. First, this thesis adapts generative incremental learning tothe problem of imitation learning. It extends the imitation learning framework to workin the multi-agent setting where observations gathered from multiple agents are used toinform the training process of a learning agent, which tracks a dynamic target. Sincedriving has associated rules, the second part of this thesis introduces a method to provideoptimal knowledge to the imitation learning agent through an active inference approach.Active inference is the selective information method gathering during prediction to increase apredictive machine learning model’s prediction performance. Finally, to address the inferencecomplexity and solve the exploration-exploitation dilemma in unobserved environments, an exploring action-oriented model is introduced by pulling together imitation learning andactive inference methods inspired by the brain learning procedure.
YR 2023
FD 2023-04
LK https://hdl.handle.net/10016/38244
UL https://hdl.handle.net/10016/38244
LA eng
NO Mención Internacional en el título de doctor
DS e-Archivo
RD 17 jul. 2024