Construction Robotics
The construction industry has a great number of challenges and even bigger potential for mobile robots to contribute to. In RSL, we are interested in solving challenging problems of construction site preparation (rock breaking, digging, material handling) as well as construction site inspections. We tackle the challenging and unsolved tasks that can be performed autonomously by a mobile platform of various sizes.
Heavy Construction
Autonomous Locomotion
In unstructured or hazardous environments, the movement of heavy construction machinery presents significant challenges in terms of operator safety, operational efficiency, and motion precision. Our research on autonomous locomotion for these machines aims to enhance the capability to reach and navigate complex workspaces for subsequent task execution, ultimately enabling the development of more sophisticated behaviors. Our Menzi Muck research platform introduces the distinctive challenge of wheeled-legged locomotion, which further expands the complexity of the locomotion problem and demands advanced control solutions. To address these challenges, we employ reinforcement learning, physics-based simulation, and advanced perception algorithms.
Contact:
Pol Eyschen ()
Earthwork and Excavation
Earthwork and, in particular, excavation, are key tasks for heavy construction machinery. Being a repetitive and fatiguing task, it is a prime target for automation in order to address safety and labour shortage on construction sites. The complicated nature of hydraulically actuated heavy machinery in contact with the ground provides challenging research topics in many core robotics fields such as control, perception and planning. In order to automate the whole excavation cycle our research addresses everything from understanding the interaction between the machine and the ground external page http://arxiv.org/abs/2510.11574 to end-to-end learned controllers for perceptive manipulation tasks external page https://arxiv.org/abs/2509.17683.
On the higher level, moving machinery on a construction site in order to complete the excavation requires substantial planning. Our research focuses on learned and conventional methods to solve these complex real-time planning problems for construction sites external page https://ieeexplore.ieee.org/abstract/document/10733983 in order to complete more complicated, long-term tasks external page https://www.science.org/doi/10.1126/scirobotics.abp9758.
Contact:
Lorenzo Terenzi ()
Lennart Werner ()
Material Handling
A common task on construction sites is handling bulk materials with specialized tools. The underactuated grippers specialized for this task offer improved efficiency and flexibility when operated by skilled operators, but present unique challenges for automatic control, particularly when paired with hydraulic actuation. We address these challenges with data-driven modeling and reinforcement learning-based control techniques, and show successful applications in end-effector positioning [https://doi.org/10.3929/ethz-b-000704791] and active throwing [https://doi.org/10.3929/ethz-b-000697483] tasks. Combining these skills with perception and planning modules, we demonstrate the world’s first autonomous bulk material management.
[https://arxiv.org/abs/2508.09003v1]
Contact:
Fang Nan [external page Google Scholar] ()
Rock Breaking
Autonomous robotic construction typically depends on uniform, prefabricated materials, limiting its ability to use the irregular resources common in natural or remote environments. Enabling usage of found materials expands the range of usable resources, reduces dependence on prefabricated components, and supports sustainable building in challenging environments.
We develop robots capable of perceiving, breaking, reshaping raw materials, and assembling them into stable, functional structures. The research integrates advanced sensing, physics-based simulation, and reinforcement-learning control to plan and execute material analysis, targeted fragmentation, and precise assembly, creating a foundation for resilient, low-impact construction on Earth and beyond.
Contact:
Marina Gouveia [external page Google Scholar] ()
Claudio Canales [external page Google Scholar] ()
Online Learning
While reinforcement learning has achieved great success in many robotic tasks, applying RL to heavy construction machinery faces major challenges due to the difficulty in modeling and simulating the multi-domain physical system involving hydraulic and mechanical dynamics. Heterogeneity and environmental dependency further complicate the problem and makes data-driven modeling also inefficient. We show the possibility of adapting a pre-trained control policy to different machines within minutes with latent-space adaptation. Leveraging techniques from model-based RL, differentiable RL, and online learning, we are also developing extremely efficient algorithms that learns to control heavy machinery, or other challenging systems such as soft robots, by online reinforcement learning.
Contact:
Fang Nan [external page Google Scholar] ()
Construction site inspection
A construction site is a highly dynamic and changing environment by its design. Autonomous localization and navigation of the mobile legged platforms are particularly challenging due to the presence of dynamic construction material, abundance of self-similar places, for example, in the case of an apartment building, and a typically feature-poor environment. In our research, we look at how to provide the most robust solution for localization and navigation to ensure consistent and persistent autonomous construction site inspection solutions.
Contact:
Olga Vysotska [external page Google Scholar] ()
Linus Kramer ()
Looking for a project?
Check available projects
