Overview
Construction is responsible for a large share of global carbon emissions, partly because it relies heavily on energy-intensive materials such as steel and concrete. Bamboo, by contrast, is a fast-growing, renewable material with high strength-to-weight ratio, but its natural variability and flexibility make it difficult to use in automated construction.
Led by Nicolas Kubail Kalousdian and Achim Menges at the Institute for Computational Design and Construction (ICD), University of Stuttgart, this project developed a robotic system for constructing with bamboo. The project ran from 2021 to 2024 and was funded by the Cyber Valley Research Fund.
Key results
- Designed and implemented a custom mobile robotic system optimised for handling and assembling bamboo elements.
- Developed a hybrid task and motion planning (TAMP) approach that combines reinforcement learning with logic-geometric programming. Reinforcement learning enabled dynamic adaptation during bending tasks, while logic-geometric programming supported structured, multi-step planning of construction sequences.
- Created a high-fidelity simulation environment to model bamboo’s mechanical behaviour and train reinforcement-learning policies before deployment on real hardware.
- Used domain randomisation and curriculum learning to improve transfer from simulation to reality, enabling the robot to work reliably with real, non-standardised bamboo.
- Demonstrated that the robotic system could reach, transport, and assemble bamboo elements, including cooperative behaviours between multiple robots.
- Built a full-scale architectural demonstrator that executed key construction tasks, showing the feasibility of robotic bamboo construction despite remaining engineering challenges.
Why it matters
The project shows how AI and robotics can enable the use of renewable, low-carbon materials in construction. By automating the handling of flexible, non-standardised elements such as bamboo, it opens a pathway to more sustainable building practices that reduce dependence on carbon-intensive materials. The methods developed for task and motion planning, simulation-to-reality transfer, and multi-robot cooperation may also inform other forms of robotic construction and fabrication.
This project produced the following peer-reviewed publications:
- N. K. Kalousdian et al., “Learning Robotic Manipulation of Natural Materials With Variable Properties for Construction Tasks”. In: IEEE Robotics and Automation Letters, Vol. 7, No. 2, April 2022, pp. 5749-5756. DOI: 10.1109/LRA.2022.3159288.
