Reinforcement Learning for Robotic Control and Optimization: Overcoming Real-World Deployment Bottlenecks

Reinforcement learning (RL) has demonstrated remarkable potential for acquiring complex robotic skills, from agile locomotion to in-hand manipulation. Yet the transition of these policies from simulation to physical systems remains deeply unreliable. The field faces a set of urgent, unsolved bottlenecks: extreme sample inefficiency makes training on hardware prohibitively expensive; the lack of rigorous safety guarantees prevents deployment in human-shared environments; policies brittlely fail under visual and dynamical distribution shifts; and sim-to-real transfer continues to demand laborious manual tuning. While traditional control theory provides stability and structure, it cannot alone address these data-driven adaptation and generalization challenges—and current RL methods have not yet closed this gap. There is now a pressing need for RL algorithms that are not only performant but also sample-efficient, safe, transferable, and amenable to real-time hardware constraints. This Research Topic focuses precisely on these critical technical barriers, moving beyond generic invocations of “complexity” to target the specific deficiencies that currently limit robot learning.This Research Topic aims to collect and disseminate principled advances that bridge the gap between RL theory and trustworthy, real-world robotic control. It seeks to foster a focused dialogue on how RL can be redesigned to yield controllers that are data-efficient, provably safe, robust to domain shift, and executable under strict real-time budgets. We emphasize methodological depth and rigorous evaluation over breadth. We invite original research, comprehensive reviews, and case studies that address one or more of the following tightly scoped themes:• Adaptive dynamic programming and decision-making in robotics;• Reinforcement learning algorithms for intelligent robotic control;• Learning-based optimization for robotic system performance;• Data-driven analysis, modeling, and integration of robotics;• Safety, reliability, and interpretability in RL-based robotic control;• Applications of RL in industrial automation, aerospace, and unmanned systems;• Hardware-in-the-loop and real-time RL implementation challenges;We invite original research articles, comprehensive reviews, case studies, and short communications on topics relevant to this Research Topic. Topic editors: Tianjiao An, Yuning Qiu, Shunchao Zhang, Yongwei Zhang Source: https://www.frontiersin.org/research-topics/81685/reinforcement-learning-for-robotic-control-and-optimization-overcoming-real-world-deployment-bottlenecks