Reinforcement learning with construction robots: A preliminary review of research areas, challenges and opportunities

The interest in the use of robots in the construction industry has been around for several decades; however, the advancement of technology for related applications has been slow. Considering that most construction robots are not fully automated and require extra guidance and instructions from the operators, an autonomous way for robots to understand how to execute specific construction tasks is needed. Reinforcement learning (RL) is a possible solution to this problem. Instead of explicitly detailing the solution to a problem, RL algorithms enable a robot to autonomously discover an optimal behavior through trial-and-error interactions with its environment. It constructs a learning model to solve various sequential decision-making problems. RL algorithms could help construction robots establish a learning process based on the feedback from the construction site and lead to an optimal strategy to finish the sequential construction work. Nowadays, on-site construction robots still require tedious work in preprogramming. Many other areas have used RL algorithms on different tasks, such as dexterous manipulation, legged locomotion, or pathing planning; thus, there is great potential for combining construction robot applications with RL algorithms. Despite these achievements, most works investigated single-agent settings only. However, many real-world applications naturally comprise multiple decision-makers that interact simultaneously, such as traffic modeling for autonomous vehicles and networking communication packages for multi-robot control. These applications have faced significant challenges when dealing with such high-dimensional environments, not to mention the challenges for the on-site construction robots. The recent development of deep learning has enabled RL methods to drive optimal policies for a sophisticated high-dimensional environment. To the best of our knowledge, there is currently no extensive survey of the applications of RL techniques within the construction industry. This study can inspire future research into how best to integrate powerful RL algorithms to achieve a higher-level autonomous control of construction robots and overcome resource planning, risk management, and logistic challenges in the industry.