Title: Artificial Intelligence Across Disciplines: Research, Innovation, and Intelligent Solutions
Editors: Dr. Subita Bhagat, Dr. A. Balamurugan, Dr. P. Krishna Kumar, and Mrs. S. Nandhini Devi
ISBN: 978-81-69857-83-3
Chapter: 13
DOI: https://doi.org/10.59646/815/13
Authors: Mrs. V. Archana, Mrs. P. Usha, Mrs. D. Bala Gayathri, and Ms. S. Khathijath Fariha
Abstract
Autonomous and adaptive intelligent systems require robust perception, dynamic decision-making, and rapid real-time policy recalibration when operating within non-stationary, safety-critical environments. This chapter examines the convergence of deep representation learning and adaptive sequential decision frameworks, focusing specifically on urban autonomous vehicle navigation under degraded sensor perception and adversarial weather dynamics. We present an end-to-end framework integrating multi-modal sensory fusion through spatio-temporal convolutional networks, deep reinforcement learning via soft actor-critic architectures, and real-time distribution-shift detection using variational latent modeling. By evaluating the system across mixed physical-simulation testbeds, we show that adaptive deep learning mitigates epistemic uncertainties and sustains safe trajectory planning where traditional rule-based controllers fail. Experimental validations underscore notable improvements in collision avoidance margins, dynamic path efficiency, and compute-budget latency. Ultimately, this chapter establishes theoretical and practical benchmarks for developing self-supervised, resilient autonomous agents capable of continuous, life-long learning and safe operational degradation across complex dynamic environments.
Keywords: Autonomous Systems, Deep Reinforcement Learning, Adaptive Perception, Spatio-Temporal Fusion, Out-of-Distribution Detection, Trajectory Planning.