XC Robotics
Contact
A range lineup in bright morning light: one of each kind of machine the brain retrofits — a ride-on mower, a compact loader, a tractor, a utility vehicle, a sweeper and a compact excavator, each with an empty seat and the same autonomy module above it — and beside them four purpose-built robots of identical build, carrying a gripper, a hedge trimmer, a sweeping brush and a spray head

Robotics at production scale

We are building autonomy for work at scale.

Repetitive, physical work that wears people down, and that is increasingly hard to staff. We supply the autonomy in two stages: first the brain, for the machines already in service; then robots of our own, built for the work those machines cannot reach.

What we build How we build it

What we build

Two stages. The brain, then the body.

Millions of machines are already in service, each with a person driving it. Fitting them with a brain is the fast route: the hardware exists, and the job is the one it was built for. But a machine that only drives cannot pick, cut or place. That work needs an arm, and no fleet of those exists to retrofit — so we build them ourselves.

The autonomy module: a compact trinocular camera and compute unit on a short mast bracket
Stage 01 Launching 2027

The brain

  • A module that turns an operator-driven machine into one that drives itself.
  • Reads the site on arrival, sets its own working area, plans its route, finishes the job.
  • Ride-ons and sweepers, loaders and tractors, carts and utility vehicles — one module across all of them, with only the actuation layer changing.
Three purpose-built working robots on a campus lawn at golden hour, all the same platform: one trimming a hedge, one sweeping a path with a brush head, one picking up litter with a gripper
Stage 02 In design

The body

  • A family of purpose-built robots on one platform, with arms and interchangeable tools.
  • Specified and verified in simulation before anything is fabricated.
  • Landscaping first, then agriculture, logistics and industrial sites.

How we build it

One automated closed loop. A task goes in. A deployable stack comes out.

Three things go in: what the place looks like, what the job is, and what compute the machine can carry. What comes out is the whole stack — compiled binaries, trained weights, tuned parameters, and the sensor and board configuration they assume — verified in simulation and ready to flash onto bare hardware.

  1. Someone recording a back garden on a phone the morning after a party, cups and bottles scattered across the patio and lawn 01 Task

    A phone walkthrough of the site, a description of the job, and the compute budget the machine must run inside. Nothing else — no survey, no CAD model, no capture rig.

  2. Nine renders of the same garden in a grid, identical in layout but each with different debris, weather and light 02 Digital cousins

    One capture becomes an open-ended family of worlds: the same kind of place, never the same twice, with layout, clutter, weather, light and season all resampled.

  3. The software stack drawn as translucent layers stacked above a solid embedded compute board: sensor drivers, a middleware mesh, a neural network graph, conventional algorithm blocks and a control loop 03 Stack design

    Agents design what actually ships, sized to the compute budget: the algorithm architecture — what is learned, what stays conventional, how large each network can afford to be — and the sensor configuration, from vision alone to vision fused with lidar or GNSS.

  4. One robot repeated across seven floating dioramas, each a different environment — patio, lawn, kitchen, driveway, wet street, slope and warehouse 04 Train

    Training runs in parallel across the whole family of worlds, end to end or as components.

  5. A simulated robot binning a bottle, with detection boxes on the remaining litter, a cleared-area overlay, a driven-path trail and inset depth and segmentation views 05 Verify

    End-to-end, closed-loop verification of the complete robotic system: sensing, decision, control and actuation exercised together across the whole scenario set.

  6. The real robot placing a bottle into a recycling bin in the same garden, now cleared, in morning light 06 Deploy

    The verified stack is flashed onto the real machine. What it captures in the field returns as new worlds.

Brought over from automotive.

  • An advanced driving system — the software, the sensors and the chips underneath them — has to be proven across far more situations than any test fleet could ever drive, and proven before those chips have even been manufactured. The industry answered by putting all three inside one closed simulation loop.
  • We built that environment for one of the leading companies in this field, whose end customers include three of the largest automakers and their tier-one suppliers.
  • Robotics has the same problem and none of the tooling. We are moving the method across.

Who we are

Started at Cambridge. Shipped in industry.

Origin
University of Cambridge
Team
Scientists and serial founders · Cambridge alumni
Experience
10+ years each in autonomy, robotics and simulation
Track record
Earlier ventures supplying major international industrial groups, backed by the University of Cambridge and leading deep-tech investors
A stone fountain framed by a college gatehouse arch, Cambridge