Question 1

What is a physics-informed neural network (PINN)?

Accepted Answer

A physics-informed neural network is a neural network whose loss function includes the residual of governing differential equations (ODEs or PDEs), so that the model learns from both sensor data and the known physics of the system. PINNs generalize from sparse data where purely data-driven models memorize.

Question 2

What problems does Astraea Intelligence solve?

Accepted Answer

Astraea closes the sim-to-real gap in robotics. Simulators use approximated contact, linearized friction, and simplified dynamics, so policies trained in simulation fail on real hardware. Astraea builds physics-informed dynamics models from your governing equations and sparse sensor data, ready to drop into MPC, RL, or state-estimation pipelines.

Question 3

Is Astraea open source?

Accepted Answer

Yes. The Astraea Core PINN engine and equation library are open source under the Apache 2.0 license. The Astraea Pro tier — automated architecture search, managed GPU training, and production deployment integrations — is commercial.

Question 4

Which equations does Astraea support?

Accepted Answer

Astraea ships validated templates for Fossen 6-DOF underwater dynamics, Cosserat rods, Navier–Stokes, advection–diffusion, heat equation, linear elasticity, and reaction–diffusion. It also supports arbitrary custom ODE/PDE systems.

Question 5

Which deployment targets are supported?

Accepted Answer

Astraea exports to CasADi for MPC, Drake, ROS 2, ONNX, and TorchScript. Trained surrogates run at 100+ Hz on GPU.

Docs

Overview

O13 · PINN Engine

O13 · Energy Fleet — HFMARL

Interactive