What is MechaSafari?
A few humans. A very large forest. Autonomous machines that are looking for them.
The rest is up to you.
This is not a simulation.
-
01
Enter
You and a small team enter an instrumented wilderness environment.
-
02
Mission
Reach extraction. Recover an object. Protect a location. Evade detection. Complete the objective.
-
03
Contact
Autonomous machines enter the field.
-
04
Adapt
They search.
You improvise. -
05
Survive
Everything is recorded. What happens next becomes part of MechaSafari.
Scenario MS-001
First Contact
Reach extraction.
Don't let them find you.
Status: In development
- Humans
- 4
- Objective
- 1
- Machines
- Multiple
- Duration
- 90min
Controlled by design
A bounded field
Every run happens inside a mapped, fenced, instrumented property. Machines are geofenced to it.
A human on every stop
Each machine has a supervising operator with a hard stop. Autonomy never means unattended.
Detection, not contact
Machines win by finding you. They are not designed or permitted to touch you.
Rules, marshals, medics
Briefed participants, trained marshals, medical cover and weather calls on every run.
Adversary classes
The machines.
Five classes of autonomous system. Real categories from real robotics, no brands and no science fiction.
-
Class H
Humanoid
Upright · Dexterous · Deliberate
The environment was built for your body. Unfortunately, it was built for theirs too.
-
Class Q
Quadruped
Fast · Stable · Persistent
Terrain that slows you may not slow it.
-
Class A
Aerial
Elevated · Wide-field · Patient
The forest looks different from above.
-
Class G
Ground
Low · Tracked · Unhurried
It doesn't need to be fast. It needs to be where you're going.
-
Class S
Sensors
Fixed · Networked · Always on
Some machines never move. They watch, and they tell the others.
Classes, not brands. MechaSafari has no affiliation with any robot manufacturer.
The safari is only the surface.
Every MechaSafari encounter is an experiment.
Humans behave unpredictably. That unpredictability is valuable.
Our proving ground is designed to expose embodied AI to adversarial real-world conditions, and to capture systematically what happens.
Humans
Motivated, creative, unpredictable.
Encounters
Instrumented runs in messy terrain.
Discoveries
Behaviour nobody planned for.
Validation
Reproduced, or it didn't happen.
Benchmarks
Comparable across platforms.
Safer machines
The point of all of it.
The human is not training data.
Human ingenuity creates intellectual value. We intend to treat it that way.
Participants will find things: machine weaknesses, strange behaviours, evasions, edge cases, strategies nobody scripted.
The easy path is to absorb all of it, silently, into somebody else's machine-learning corpus. We intend to do the opposite. Identify the discoveries that matter. Validate them. Catalog them. Credit the people who made them, and share the value they create.
- 01Identify
- 02Validate
- 03Catalog
- 04Attribute
- 05Share value
Prototype records Illustrative only. No encounters have run yet; these show the format, not findings.
Group-split target prioritisation conflict
Low-sun thermal contrast collapse
Bring us something you think can't be broken.
Robotics teams spend enormous effort teaching machines to succeed. MechaSafari is being designed to discover, systematically, how they fail.
- 01Adversarial field testing
- 02Navigation robustness
- 03Perception edge cases
- 04Human-machine interaction
- 05Fail-safe validation
- 06Containment testing
- 07Cross-platform benchmarking
- 08Environmental robustness
- 09Recovery behaviour
- 10Independent reproducibility
This is what we're building, not a list of things we've done. No certifications, datasets, customers or partnerships are claimed here. Early partners help write the test protocols.
The media engine
Every run tells a story.
Every encounter is filmed from both sides: the humans, the machines, the air, the heat and the data. The experience produces its own story.
Concept feeds. Imagery is rendered and telemetry is simulated; no runs have happened yet.
- Episodes
- Documentaries
- Creator events
- Challenge formats
- Technical breakdowns
Founding principles
- Humans should understand the machines entering their world.
- Capability deserves scrutiny.
- Failure is information.
- Human ingenuity has value.
- Autonomy requires accountability.
- The ability to stop a machine matters as much as the ability to start one.
We are fascinated by robotics. We are not Luddites. We think extraordinary machines deserve extraordinary testing.
Measure capability. Expose failure. Validate safety. Preserve human agency.
Read the principlesInterest register
Join the first hunt.
No dates, no prices, no countdown. Tell us who you are and we'll tell you when this becomes real.
Origin: 08:17
MechaSafari began as an objectively ridiculous text conversation about hunting robots in the woods.
At 08:24, mechasafari.com was registered for $11.48.
The joke lasted approximately ten minutes.
Then we realised the testing environment might be more valuable than the robots.