01
Robot data engines
Collection, curation and feedback loops that turn operating hours into training signal. Data is the scarcest input in the field and the hardest to buy.
Portfolio
No investments yet. Here is the map instead.
Fund I has not closed, so there is no portfolio to show and we will not dress one up. Publishing the constraint map is the more useful disclosure — it is what we should be judged against later.
The map
Seven families we underwrite.
Each is a place where the field repeatedly fails in the same way. Leverage — whether one solution can serve many systems — is the whole argument, so we rate it explicitly rather than implying it.
02
Simulation & sim-to-real
Scenario generation, domain randomisation and transfer measurement. The gap between simulated and real performance is still paid for in hardware and time.
03
Evaluation, safety & observability
Benchmarks, regression harnesses, incident analysis and runtime monitoring. Capability is advancing faster than the ability to say whether it works.
04
Edge inference & orchestration
Running competent models on the robot within latency, power and reliability budgets, and coordinating them across a machine.
05
Manipulation primitives
Reusable grasping, force control and contact-rich skills that transfer between platforms rather than being rebuilt per robot.
06
Fleet & intervention infrastructure
Deployment, teleoperation, escalation and utilisation tooling. The layer that turns a successful pilot into an operation.
07
Selective enabling hardware
Tactile sensing, force control and actuation where the physics genuinely gates the software layer. Selective, because most hardware is not a constraint.
No families match that filter.
Limits
The concentration rules we intend to hold ourselves to.
A constraint fund fails by drifting: into adjacent categories, into one vertical, or into a single platform's ecosystem. These are the limits that keep the construction honest.
- Single company, of committed capital Max 10%
- Single constraint family Max 30%
- Single end vertical Max 35%
- Dependent on one platform vendor Max 25%
- Selective enabling hardware Max 20%
- Reserved for follow-on ~45%
Working on one of these seven?
We are pre-portfolio, which means the first cheques are still unallocated. If you own one of these constraints, we want to hear about it early.