UniSim Unified API

UniSim is the sole interface layer between FastSim and the underlying simulation engines. It abstracts "start the simulator, advance physics, query running state" into a fixed set of interfaces, so SceneManager, ControllerManager, and other upper modules are completely unaware of which backend is running.

Core Interfaces

Method	When Called	Description
`initialize_simulation()`	During `setup_all_before_loop()`	Start the simulation App/SimulationContext
`setup_viewport_camera()`	After init (optional)	Set default viewport for debugging
`reset()`	After scene construction	Reset physics, run initialization steps
`step()`	Each simulation frame	Advance physics by one timestep (`dt`)
`is_running()`	Main loop condition	Query whether the backend is still running

Position in the Simulation Loop

text

ControllerManager.run_all_commands()     ← Execute command queue
SceneManager.pre_physics_update()        ← Write target values to simulator
UniSim.step()                            ← Physics engine advances one frame
SceneManager.post_physics_update()       ← Read latest state from simulator
[step callbacks]                         ← Extension hook points

Backend Selection

The backend is determined by simulation.stereotype. UniSimFactory.create(config) instantiates the corresponding backend:

yaml

simulation:
  stereotype: isaaclab   # or mujoco, pybullet
  dt: 0.01667

IsaacLab Backend

GPU-accelerated physics and rendering via NVIDIA Isaac Lab. Supports cuda:0 device, PhysX GPU pipeline, and high-fidelity rendering.

Source: fastsim/simulators/isaaclab/

MuJoCo Backend

Lightweight CPU physics engine. Fast startup, suitable for debugging and rapid iteration. Supports MJCF and URDF asset formats.

yaml

simulation:
  stereotype: mujoco
  dt: 0.002

Source: fastsim/simulators/mujuco/

PyBullet Backend

Open-source CPU physics engine with built-in OpenGL rendering. Commonly used in reinforcement learning research.

yaml

simulation:
  stereotype: pybullet
  dt: 0.002

Source: fastsim/simulators/pybullet/

Custom Backend

To integrate a new simulator, inherit UniSim and implement the five interface methods, then register via stereotype:

python

from fastsim.unisim.unisim import UniSim
from fastsim.annotations.stereotype import stereotype

@stereotype("my_simulator")
class MySimulator(UniSim):
    def initialize_simulation(self): ...
    def step(self): ...
    def reset(self): ...
    def is_running(self) -> bool: ...