Where we stand
We are fascinated by robotics. We are not Luddites.
The machines arriving now are extraordinary, and they will share the world with everyone, not only with the people who built them. We think extraordinary machines deserve extraordinary testing, and that the rest of us deserve to see how they behave when things get messy.
Six principles
Humans should understand the machines entering their world.
Most people will meet autonomous machines without ever having seen one fail. MechaSafari makes that knowledge visible: to participants, to viewers, and to the teams who build them.
Capability deserves scrutiny.
Demos are recorded under ideal conditions. Scrutiny happens in the rain, at dusk, with someone actively trying to make it go wrong.
Failure is information.
A machine that fails in our field is a machine that didn't fail somewhere worse. Every failure is logged, reproduced if it can be, and reported.
Human ingenuity has value.
When a person discovers something new about how a machine behaves, that discovery came from them. We intend to identify it, validate it, credit it and share the value it creates.
Autonomy requires accountability.
Every machine in the field has a named human responsible for it, a hard stop, and a record of what it did.
The ability to stop a machine matters as much as the ability to start one.
Stop authority is tested as seriously as any capability. If a machine can't be stopped reliably, it doesn't run.
The human is not training data.
Every interaction between a person and a machine produces data. Most of it is ordinary. Some of it is a discovery: a failure nobody knew about, a behaviour nobody designed, a strategy that works across platforms.
We intend to treat those discoveries as intellectual work with an author. A discovery has to be novel, reproducible and meaningful. When it is, it is catalogued with the name of the person or team who found it.
How the value is shared (credit, compensation, a share of licensing) is still being designed. We will publish the mechanism before the first run, not after.
What we will not build
- A weapons-training range.
- Anything designed to make machines better at harming people.
- A spectacle that depends on someone getting hurt.
The adversarial relationship is the product. Violence is not.