Search papers, labs, and topics across Lattice.
Deep reinforcement learning (DRL) has emerged as a transformative paradigm for developing autonomous systems and robotic agents capable of acquiring complex behaviors through interaction with their environments. The rapid proliferation of DRL techniques across domains such as navigation, manipulation, multi-agent coordination, and safety-critical control necessitates a systematic synthesis of the existing body of knowledge. Our objective in this systematic literature review is to provide a comprehensive and structured analysis of DRL methodologies applied to autonomous systems and robotics, identifying principal research trends, methodological evolutions, and persistent challenges. To conduct this review, we followed a rigorous systematic methodology encompassing a predefined search strategy, explicit inclusion and exclusion criteria, and a multi-stage screening process to ensure the relevance and quality of the selected studies. The extracted data were organized along seven key dimensions: foundational frameworks, autonomous navigation and path planning, robotic manipulation and physical interaction, multi-agent systems and swarm intelligence, safety and verification, advanced learning paradigms, and domain-specific applications. Our results reveal a clear trajectory from foundational algorithms toward hybrid approaches that integrate model-based planning, hierarchical learning, and constrained optimization. We further observe that safety verification and robustness remain critically underexplored relative to performance-oriented advancements, and that distributed learning in multi-agent settings presents unique convergence challenges. From these findings, we conclude that the future of DRL in autonomous systems lies in the development of sample-efficient, verifiably safe, and generalizable algorithms. This review provides researchers and practitioners with a structured map of the current landscape, guiding future work toward closing the gap between simulated success and real-world deployment.
