Deep Agency

Human Guided Deep Reinforcement Learning for Wood Structure Assembly

ITECH Master Thesis 2023, in collaboration with Sarvenaz Sardari, Selin Sevim
University of Stuttgart

Traditional timber joint assembly task is hard to automate.
Because every piece of wood has different fiber grain, and its shape can change with humidity.

Can robot have human-like intuition through deep learning?

An artificial neural network can learn skills such as playing Go game or controlling robots by interacting with its environment.

An artificial neural network can learn skills such as playing Go game or controlling robots by interacting with its environment.

Deep Agency is a robotic training and control workflow for adaptive robotic assembly for wood joints

Customized reinforcement learning environment in simulation allows the neural network to explore and improve its skills

Human demonstration on the real world provide data for the agent to learn more efficiently

The trained neural network can be deployed on the real robot through a customized light weight real-time inference control pipeline

More details coming soon…

Next

Glass Vault