Suspended Manipulator


Our journey in agriculture started with floating robot. This was in response to the growing demand for labor in harvest season. The concept was designed initially to tackle sloped farms where existing tractors cannot operate, but it offers more:

-    Runs on solar power without CO2 emission
-    Doesn't compact the soil
-    Enables for more efficient use of the land
-    Covers large areas
-    Manipulates objects and transports goods

Possible revolution ...

Farmers rely on tractors heavily. Some tasks are rapid and well suited for these machines, yet many can be done over time. Our robot is slower, but works all day long on solar power, also over night with extra batteries. At Floating Robotics, we develop various farming skills for robots. We believe precise manipulation in combination with the floating robot is a straight way to sustainable high-value crop production.

-    Reduce CO2 emissions
-    Save water by precise irrigation
-    Reduce  pesticides & herbicides health risks for farmers
-    Make horizontal farming competitive again

Floating robot is not a tool, but a skillful farm assistant.

Radical design

Floating manipulator runs on cables above the plants. In comparison:

-    Drones shake the plants and operate shortly
-    Wheels compress the soil and slip on slopes
-    Legged robots are power-hungry and expensive

With four actuators only, Floaters stand on top of the list in terms of weight, energy, and simplicity of navigation. They are most suitable for repeated tasks in large but limited work spaces.

What's novel?

Cable-driven robots usually put winches outside to save weight. Our robot carries them onboard together with computers, batteries, and all other electronics. This makes it compact , portable, and easy to deploy. We have a pending patent on mechanical coupling of actuators to cut energy consumption by 50%. Floating robot can locate itself in any arbitrary installation using the onboard sensory equipment. Actuators can perform position and force control which provide precision or compliance respectively.


Cables are taken out of Floaters like a vacuum cleaner and installed within minutes. In outdoor environments, especially sloped farms, we install four poles at the corners of the field. In the case of larger areas, a grid of poles can be installed where the robot should be manually moved when it finishes tasks within one sub-grid. Floaters can also work inside structured areas like tunnel greenhouses, co-working spaces, warehouses, big halls, etc.


To move around in the field, a high-level position controller is used. It searches for a static stable orientation with a nonlinear optimization given a desired base position and known pole positions. Based on the orientation of the robot, cable lengths and forces can be calculated. These quantities are provided to the lower-level motor controller firmware. The orientation of the robot is not controllable since it is underacted and is therefore also an optimization variable.


Suspended robot is able to obtain a field map pretty easy just by moving above the plants. Lidars can obtain raw point-clouds from multiple canopies at the same time. We have developed an algorithm to separate canopies by clustering the point clouds. This helps the robot build a map and use it online for other tasks.


Perturbation of wind is rejected with passive dampers in the corners and simple active control laws that feedback the IMU angular velocities to cable lengths. Passive dampers take out large perturbations while the active control law brings the robot to full stability. The robot also operates in pure force mode and shows significant compliance and stability.


Using stereo vision and deep learning, the robot is able to detect crops and calculate a picking point. This cycle is continuously repeated until cameras don't see new objects. Grapes are dropped in a basket currently, but could be loaded onto the robot to unload at a corner.


Using a simple mapping algorithm and absolute positioning the robot has a complete understanding of the environment. With this information it can follow canopies and perform continuous tasks. In this case, the manipulator is carrying a trimmer and following a line at a certain height for pre-pruning the plants in winter. The manipulator can be further programmed to perform precise pruning.

Yield estimation

The robot is able to estimate the overall weight of all the crops it sees with an advanced camera. This feature comes after extensive data collection and precise laboratory measurements. The algorithm is programmed to be robust to viewing perspective, distance, and the number of crops observed.

Disease detection

Moving cameras around allows the robot to obtain a map of disease intensities and frequencies. We've trained neural networks to detect Mildew spots and Esca for now, but the approach is expandable to deficiencies too. Although it's already late to treat when human eyes or RGB cameras can see the phenomena, the Suspended Manipulator provides the opportunity to scan the entire farm closely with other sensor types, e.g. multi-spectral.


This project was funded by Innosuisse grant 52803.1 IP-ICT in 2021 and 2022. We continue our journey in tomato greenhouses by commercializing the harvesting system.

Your Heading