Engineers Turn a Petrol Two-Wheel Tractor Into an Electric, Weed-Sensing Farm Machine

A quiet revolution is rolling across the world’s smallest farms, and it started in a university workshop with a humble petrol-powered tiller. Engineers at the University of Sydney have systematically converted a commercially available Grillo G55 two-wheel tractor into an electric machine, keeping its familiar gearbox, power take-off and handlebar controls intact while swapping the combustion engine for a battery-driven electric motor. The work, published in Smart Agricultural Technology, forms part of an Australia–Cambodia collaborative research project funded by the Australian Centre for International Agricultural Research, and it offers something the emerging electric-tractor literature has largely lacked: a documented, workshop-reproducible recipe that ordinary technicians can follow.

Two-wheel tractors are the workhorses of smallholder agriculture across sub-Saharan Africa and Southeast Asia. They are affordable, versatile enough to haul, pump water and process harvests as well as till fields, and well suited to the small, irregular plots that dominate smallholder landscapes. Yet nearly all of them run on petrol or diesel, bringing recurring fuel costs, particulate emissions and noise levels at the operator’s ear that previous field studies have flagged as a long-term health concern. In Cambodia, where two-wheel tractors are ubiquitous, they are also frequently involved in crashes linked to poor visibility and limited driver experience with motorised traffic. For poorer households and for women farmers, who can face additional barriers to operating and maintaining petrol machinery, these factors can be decisive obstacles to adoption.

The Sydney team’s answer was not to design a new machine from scratch but to electrify an existing one. They stripped out the Loncin G200FA petrol engine, its mechanical clutch, fuel tank, air filter, exhaust and ignition system, and replaced them with a 1 kW brushless DC motor rated at 24 volts and 3,000 rpm, delivering 3.3 newton-metres of torque. Because an electric motor does not idle, the clutch became unnecessary; the motor now drives the original gearbox directly through a custom-machined face plate and a three-piece coupling that connects to the gearbox’s spline shaft. Crucially, no gearbox internals were modified, making the conversion reversible and modular. The gearbox retains its two forward and two reverse gears, and the power take-off continues to deliver its fixed 965 rpm for implements.

The electrical architecture is deliberately simple and maintainable. Two 12-volt, 40 amp-hour lithium iron phosphate batteries wired in series supply the motor through a Roboteq controller capable of handling up to 80 amps continuously. The controller reads throttle and safety inputs while processing encoder and Hall-effect feedback from the motor for commutation and torque control. Wiring is organised into four modular regions: handlebar controls, a fairing housing safety devices, a routing board for cable management, and the tractor body containing the power components. Safety features include a repurposed clutch lever acting as a dead-man switch, an emergency stop with a contactor, and a beacon that lights when the tractor is armed. If the operator releases the dead-man lever or presses the emergency stop, power is cut instantly and the tractor halts, with the throttle required to return to zero before movement can resume.

Performance figures from laboratory and field testing suggest the conversion is more than a curiosity. Assuming 80 percent mechanical efficiency, the motor’s 3.3 newton-metres translates through the gearbox ratios of 250.7:1 and 106.1:1 into theoretical wheel-shaft torques of 661.8 and 280.1 newton-metres respectively, with linear speeds of roughly 0.9 and 2.1 kilometres per hour. Workshop trials found 2.1 kilometres per hour slightly faster than a comfortable walking pace when pulling a plough, so the practical maximum was capped at 1.8. The team measured pull forces of about 40 kilograms-force on a smooth laboratory floor and at least 50 kilograms-force in the field during wheel slip, and observed that the machine copes easily with a 25-degree incline and climbs obstacles nearly 20 centimetres high, about half its wheel diameter. Operating duration is estimated at more than 1.5 hours under realistic conditions, a conservative figure carrying roughly a 50 percent safety margin, with around 56 minutes possible at continuous peak power.

Initial outdoor trials at the university’s Cobbitty field site put the prototype to work on sandy loam with a plough attached at a working depth of 5 to 7 centimetres. Wheel encoders tracked the driven wheels while a 720-pixel webcam estimated true ground speed using optical flow, revealing the slip between them. Across 2,559 logged frames in one trial, mean encoder speed was 0.208 metres per second against a camera-estimated ground speed of 0.203, yielding a median slip ratio of 3.06 percent. That persistent, non-zero slip under drawbar load is exactly the kind of quantitative baseline the team says will underpin future traction models and control strategies, though they caution the measurements are preliminary and limited to soft soils.

The most eye-catching element of the project is its first step toward smart mechanisation: a laboratory-tested Green-on-Brown smart weeding workflow. Using the open-source OpenWeedLocator detection method running on a Raspberry Pi 5, the system identifies green vegetation against brown soil, divides each frame into left, centre and right spray zones, and triggers relay-based outputs for each zone. Detections count as in-band when their centres fall within the lower 30 percent of the frame, nearest the expected spraying area, with a minimum activation time of 0.25 seconds to prevent rapid switching. On recorded video at 640 by 480 resolution, the desktop version processed frames at an effective rate of 52.54 frames per second, while the Raspberry Pi managed 40.67, with broadly similar zone-activation behaviour. The team stresses this is a proof of concept only: no field validation of weed detection, spray efficacy or relay response has yet been performed.

To place their approach in context, the researchers compared their converted machine with AfTrak, a purpose-built solar micro-electric tractor developed for Deep Bed Farming in Malawi. AfTrak pairs a custom walk-behind unit with a 3-kilowatt-peak solar base station that also supplies zero-carbon electricity for lighting, phone charging and clean cooking, and is specified for soil preparation of 10 square metres per hour at 300 to 400 millimetre depth. The Sydney e2WT, by contrast, is currently charged from laboratory power, with portable solar arrays, shared community charging hubs and integration with existing local solar systems identified as future options. The authors frame the two not as rivals but as complementary pathways: purpose-built integrated systems on one side, and incremental conversions that preserve the familiar mechanical layout, implements and operator skills farmers already have on the other.

The conversion was designed from the outset for replication. Custom machining was limited to a handful of parts producible with basic milling and drilling, off-the-shelf components were sourced through standard industrial suppliers, and the bill of materials, roughly 9,000 Australian dollars including the base tractor and tiller, plus an estimated 13 weeks of full-time-equivalent labour, is documented alongside the tools required. This transparency is aimed squarely at the Cambodian partners, who are undertaking parallel platform development with the Royal University of Agriculture, the Institute of Technology of Cambodia and Engineers Without Borders Australia, and who can now skip many of the design dead ends the Sydney team encountered, from weight-distribution shifts caused by battery placement to cable-routing and throttle-response quirks.

Limitations remain, and the authors are candid about them. Operating duration rests on estimates rather than measured battery depletion; traction and weight distribution need broader evaluation across soils and implement loads; camera-based speed sensing can be confounded by low-texture soil, changing light and vibration; and no locally parameterised cost or life-cycle comparison with the petrol original has yet been made. Still, the prototype demonstrates that electrifying the machines smallholders already own is practical, safe and increasingly smart. If solar charging matures for tropical conditions, the sight and sound of a silent, fume-free tractor weeding its own way down a Cambodian paddy may not be far off.

Subject of Research: Electric conversion of a commercial two-wheel tractor for smallholder farming with preliminary smart weeding integration

Article Title: A systematic conversion of a two-wheel tractor to electric power with preliminary smart mechanisation for smallholder farming

Article References: Farhood, H., Muller, E., Tsoukalas, G., Al Matar, F., & Sukkarieh, S. (2026). A systematic conversion of a two-wheel tractor to electric power with preliminary smart mechanisation for smallholder farming. Smart Agricultural Technology, 15, Article 102608. https://doi.org/10.1016/j.atech.2026.102608

Image Credits: AI Generated

DOI: 10.1016/j.atech.2026.102608

Keywords: electric tractor, two-wheel tractor, smallholder farming, agricultural mechanisation, brushless DC motor, Green-on-Brown weeding, OpenWeedLocator, Raspberry Pi, wheel slip, Cambodia, solar charging, precision agriculture