Arm Create 2026 / Physical AI track

AEOLUS
simulation interface

A safety-governed Physical AI simulation with a measured Arm optimization: the deadline-era FP32 forecast artifact uses 50% less model-array memory and runs 1.73× faster than FP64 on native Neoverse-N2, while preserving prediction parity.

Deadline-era optimization output

One workload. Native Arm64. Measured before and after.

FP64 and FP32 ran the same batch-one forecast on an Arm Neoverse-N2 runner. The reduced-precision artifact passed a predeclared prediction-parity gate. These are development benchmark results, not power, NPU, deployment, or qualification claims.

1.73×median speed-up738.349 → 426.358 µs
50%less array memory28,759,024 → 14,379,512 bytes
47.1%smaller artifact2,126,337 → 1,124,273 bytes
4.88×10⁻⁶maximum normalized driftpassed the 1×10⁻⁴ parity gate

What we are building

Not a dashboard about a simulation. The interface for the simulation itself.

AEOLUS models a distributed habitat ventilation system. Rooms and processing areas are connected by directed actuators. The Connections view lets the team define that topology before it is consumed by the Python plant model.

The current learned adviser forecasts future telemetry under each candidate action. A deterministic Habitat Management Computer reviews every proposal and remains the sole actuator authority. The evidence layers make forecasts, arbitration, plant response, and replay visible without exposing hidden simulator truth to the model.

Editable layerConnectionsEvidence layersLive system / Scenarios / Telemetry / Benchmarks
01Define roomsCabins, labs, processing bays
02Connect actuatorsOne-way or paired directions
03Run the plantTelemetry and fault scenarios
04Show recoveryBounded virtual action

Project resources

Follow the work where it is actually happening.

Team / scope

Built by Alex, Ben, and MS-Mesh for the Arm Create challenge.

Simulation only. No live plant, production telemetry, or real actuator command is involved. Abstract values in the current model are not spacecraft measurements or safety thresholds.

Read the current repository README
AEOLUS / Arm Create 2026Open-source developmentPhysical AI track