Farm robots could become more affordable for smaller producers by reusing batteries and motors from retired electric vehicles, according to a prototype that cut the cost of its frame and driving components by around 60%.
Researchers in China used second-life parts to create an adjustable robot that uses artificial intelligence to distinguish crops from weeds, which can then be treated with a laser rather than herbicides.
The reused components cost less than US$450, and the developers say the approach could give smallholders access to precision technology generally developed for larger farms.
Reusing electric-vehicle components
The TS-MAMP prototype was developed by researchers from West Anhui University, Nanjing University of Information Science and Technology and Westlake University.
In findings published as an arXiv preprint, they explained that they recovered two 48 V motors from retired low-speed electric vehicles and tested them to ensure they performed similarly.
They also selected used lead-acid batteries that retained between 60% and 80% of their original capacity. Four were connected to create the robot’s battery pack.
Because used batteries deteriorate at different rates, a balancing system transfers energy between them during charging and operation. Chains and gears increase the turning force reaching the wheels, reducing strain on the older motors.
The researchers estimated that using second-life parts cut the cost of the frame and driving system by around 60% compared with buying equivalent components new.
However, the $450 figure excludes the camera, onboard computer and laser-weeding equipment. The researchers have not calculated the cost of a complete machine or assessed the environmental savings from reusing the components.
Interested in farm robotics? You might like these articles
Adjustable design for different crops
The robot was designed for small and fragmented farms, where crop types and row widths vary and large autonomous machines may be difficult to justify.
Its wheel spacing can be adjusted from 1.2 m to 2 m, while equipment fitted beneath the frame can also be repositioned.
Tests showed that the frame could support a stationary load of at least 200 kg and tools could be changed in five minutes.
The researchers trained the weed-recognition system using 675 field images of young pak choi alongside four common weeds: barnyard grass, goosegrass, crabgrass and common purslane. Distinguishing between them can be difficult because young crop and weed plants can look similar.
The AI was trained using images with different brightness, colours and orientations. The researchers also added 500 images containing none of the target plants, helping it learn when not to identify something as a crop or weed.
In testing, 83.31% of the plants identified by the system were classified correctly, while it detected 78.41% of those present. Its overall weed-recognition score reached 80.87%, up from 62.15% before the training process was improved.
The developers said the system runs on a small computer carried by the robot, allowing it to analyse images without a reliable internet connection.
Laser performance
The prototype can carry a 50 W laser directed towards weeds identified by its camera, potentially controlling them without herbicides or disturbing the soil. A safety system prevents the laser from firing if the robot travels at less than 0.1 m/second or cannot identify its target accurately.
Field tests showed that the robot could move using the reused components, carry its equipment and run the weed-recognition system onboard.
However, the paper does not report how many weeds the laser killed, how much land the robot could cover or how long its batteries lasted. The recognition system has also only been tested on pak choi and four weed species in one region of China.
Although larger trials are needed, the researchers said the design shows how second-life electric-vehicle components could provide a lower-cost route into precision farming for smallholders currently priced out of commercial automation.
Key takeaways
- The prototype uses batteries and motors recovered from retired low-speed electric vehicles.
- Its frame and driving components cost less than US$450 – around 60% less than equivalent new parts.
- The quoted cost excludes the camera, onboard computer and laser-weeding equipment.
- The robot uses AI to distinguish young pak choi from four common weed species.
- Its overall weed-recognition score reached 80.87%, up from 62.15% before training improvements.
- Further trials are needed to assess weed control, battery runtime and the total cost of a complete machine.
Want to read more stories like this? Sign up to our newsletter for bi-weekly updates on sustainable farming and agtech innovation.
