DescriptionUnlike construction of buildings above ground where materials and external influences are well known, tunnel construction involves uncertainty and dealing with a geological environment. Driven by experience, blasting, cutting, and drilling are some of the methods that were developed throughout centuries to excavate tunnels. Consequently, different tunneling paradigms evolved and further development is sometimes hindered by conservatism.
The process of conventional tunneling however, shows many similarities to reinforcement learning, where a geotechnician (agent) tries to achieve a breakthrough (goal) by excavating the rockmass (environment) in an optimum way. We therefore designed the TunnRL (Tunnel automation with Reinforcement Learning) framework, which consists of a virtual tunneling environment and an agent that learns to achieve a breakthrough. Results show that the found policies have great similarities to current practices of sequential tunneling and the framework has the potential to discover new tunneling strategies.
|Period||28 Jun 2021|
|Event title||Reinforcement Learning Industry Meetup #2|
|Degree of Recognition||National|