Discovery & concept development
Physical AI, built from real needs.
We’re developing the foundations of a modular mobile robot. Agriculture is our first application: we’re starting with growers’ experience, a small platform, and one carefully defined task.
- Project
- Modular mobile robot
- First application
- Outdoor weeding & disease control
- Near-term goal
- A testable, limited-scope prototype
A useful robot starts with a good question.
Real-world work combines repetitive tasks with delicate decisions. A useful robot has to fit the task, the space, and the people working around it.
Fieldwork at Hoffman’s Strawberries pointed us to two needs the farm’s owner described as real pain points: weeding and crop disease control. We’ll start testing outdoors, in the same conditions as on-site trials at the farm. The final task will be shaped by grower input and early testing.
What actually takes time?
Understand recurring tasks, current tools, and the labour involved in getting the job done.
What does a good result look like?
Define what good weed removal and disease spotting look like, and when a task needs human judgement.
What fits the space?
Measure working heights, access, obstacles, and the routines the robot would need to work within.
A focused path to a first prototype.
A four-month development goal, organised around learning milestones. Timing depends on access, resources, and test results.
stage 1/4 discover
01 Discover
Listen & define
Visit farms, record the current workflow, and choose a bounded task with a clear baseline.
Current focus
02 Design
Build & bench-test
Develop the platform and task module. Measure positioning, repeatability, and tool behaviour.
Planned
03 Integrate
Connect the system
Bring sensing, controls, and hardware together in a limited, repeatable test environment.
Planned
04 Evaluate
Test & learn
Compare supervised robot operation with the existing workflow and document what still needs work.
Planned
One system. Four connected disciplines.
A modular architecture lets us test each part, learn from it, and bring the pieces together.
- MechanicalMobile platformA compact base designed around the terrain, crop access, and stability required by the chosen task.
- HardwarePower & interfacesMotor control, power distribution, sensor connections, and dependable low-level operation.
- PerceptionVisual understandingCamera-based observation and target localisation, tested against real lighting and occlusion.
- ControlMovement & task controlPositioning, controlled motion, and coordination between the platform and a task-specific tool.
Research direction, not a product claim.
Autonomy, task completion, labour savings, and real-world benefits will need to be measured. Any disease-control function depends on an appropriate agronomic protocol.
Our classroom extends to the field.
On our October 23 visit to Hoffman’s Strawberries, the owner walked us through the work behind the harvest. Weeding and crop disease control came up as everyday challenges, and they now shape our first task.
Read the visit recap ↗Progress you can put to the test.
A successful demonstration is a starting point. We want to understand how the system performs across repeated tasks, how often it needs help, and whether it improves the overall workflow.
- Task completion
- Time per target
- Human intervention
- Handling quality
- Maintenance time
- Net labour input
Have a challenge we should understand?
We’d love to hear from growers, companies, researchers, and mentors who can help turn a promising idea into a useful first build.