Historobot

Histopathology lab automation

Autonomous robotic skills for every histology lab.

We build autonomous, non-deterministic skills that let robots do the work of a histotechnologist, from tissue grossing to staining. They adapt to what they see instead of replaying a script. Any robot, any vendor, any protocol.

Simulation · 8× speed

Approach

Software at the core.

Labs already run instruments from many vendors, at every level of automation. We deliver working robotic systems, and what we build is the software inside them: skills that adapt to each lab's instruments, layout and protocols, so the lab doesn't have to adapt to the machine.

  • One skill set, many robots

    Universal histotechnical skills, trained to carry across arms, grippers and workcells.

  • Fits the lab you have

    Adapts to your level of automation, your mix of vendors and your protocols.

  • Autonomous, reliable, affordable

    The goal is full autonomy at the bench, at a cost a routine lab can justify.

  1. Grossing
  2. Cassetting
  3. Processing
  4. Embedding
  5. Sectioning
  6. Staining
Histotechnical procedures, annotated

Data

The largest dataset of histotechnical procedures.

We are collecting 50,000 hours of histotechnical procedures, recorded at the bench in hundreds of laboratories around the world. Our foundation model is trained on it, and nobody else has it.

Adaptation

Show it once.

A new lab, instrument or protocol should not mean a new training project. Our system adapts to a new environment from a single demonstration. That is how you train a person.

Compliance

GLP-compliant by design.

A robot can account for its work in a way a busy bench cannot. Every movement is tracked and written to an audit trail as it happens, tied to the specimen, the protocol and the video of the action.

  • AttributableEach action names the specimen, the station, the protocol revision and the policy version that performed it.
  • ContemporaneousRecords are written at the moment of the action, with a UTC timestamp.
  • OriginalEntries are chained by hash and linked to the camera footage, so later changes show.
  • AccurateEvery step ends in a checked outcome, not an assumption.
{
  "record":    "audit-trail/v1",
  "specimen":  "S26-04817",
  "block":     "A3",
  "cassette":  "C-7731902",
  "protocol":  "SOP-GR-012 rev 4",
  "station":   "bench-02 / dual-arm",
  "policy":    "histo-skills 0.9.2",
  "events": [
    { "t": "2026-10-10T09:14:02.180Z", "step": "grasp_slice",
      "tool": "forceps", "result": "ok" },
    { "t": "2026-10-10T09:14:03.760Z", "step": "carry_slice_to_cassette",
      "tool": "forceps", "path_mm": 96.4, "result": "ok" },
    { "t": "2026-10-10T09:14:04.960Z", "step": "seat_slice_in_cassette",
      "tool": "forceps", "result": "ok" },
    { "t": "2026-10-10T09:14:06.300Z", "step": "verify",
      "check": "piece_at_destination", "result": "pass" }
  ],
  "video":     "sha256:9f2c…e41a",
  "previous":  "sha256:51b7…0c3d",
  "reviewed":  null
}
Example record · illustrative values

Current progress

From a technologist's hands to a robot.

These are early results. Each starts from a video of a person working at the bench. We track the tools and the tissue, then reproduce the task with two robot arms in simulation.

Bench video · annotated

Annotate a technologist's hands

Forceps, tissue and cassette are tracked frame by frame, and the task is split into named steps.
Simulation · 4× speed · four cameras

Transfer it to the robot

The same pick and place in simulation, following the path the technologist's forceps took.

Contact

Talk to us.

We are happy to chat if you want to learn more.