DOCS_CORE_v4
REF_0x7E2A

Developer Portal.

complete documentation for the iacon robotics sdk. architecture of high-fidelity differentiable physics, causal behavioral models, and real-time hardware orchestration.

01_SYSTEM_INSTALLATION

Python SDK

primary research interface. support for pytorch, jax, and high-performance tensor manipulation.

pip install iacon-sdk

C++ Runtime

bare-metal deployment for edge hardware. zero-std compatible for microcontroller integration.

git clone iacon-core-cpp
DOCKER_IMAGE // GPU_OPT
STABLE
1# For NVIDIA/CUDA accelerated simulation
2
$ docker pull iacon/runtime:v0.4.2-cuda12.4
QUICK_START_READY

initialize your first bipedal gait simulation in under 60 seconds.

02_CORE_ARCHITECTURES

ARK_KERNEL_OS_v4.0
APPLICATION_LAYER
ARK_LBM_KERNEL
KINEMATIC_SOLVER
HARDWARE_HAL_LAYER
a deterministic execution stack that decouples neural reasoning from physical actuation. the kernel ensures millisecond-exact control loops while the LBM handles high-dimensional state abstraction.

Kinematic Manifolds

Differential geometry applied to robot joints. Resolves singularites and joint limits in closed-form tensors.

Causal LBMs

Large Behavior Models treat proprioception as a token stream, allowing cross-morphology knowledge transfer.

Parallel Sim

GPU-native physics kernels. Execute thousands of concurrent worlds without CPU-GPU synchronization overhead.

03_API_REFERENCE // PHYSICS

MODULE: iacon.physicsSTABLE_v4
World(config)RETURNS: PhysicsWorld

Initialize a hyper-parallel simulation container. Handles GPU memory allocation for N parallel environments.

step(actions)RETURNS: StateTensor

Execute a single time-step across all parallel environments. Synchronizes joint targets and collision states.

reset(env_ids)RETURNS: None

Resets the specified environment indices to their initial URDF configuration with randomization.

get_jacobian()RETURNS: Tensor[N, DOF, 6]

Compute the end-effector Jacobian matrices for all environments in a single differentiable pass.

MODULE: iacon.lbm

transformer-based inference engine. optimizes token sequence processing for real-time motor control.

MODULE: iacon.telemetry

high-frequency signal logging and reality-gap analysis tools.

import iacon.physics as physics

# Load bipedal morphology
robot = physics.load_urdf("models/biped_v1.urdf")

# Initialize 4096 parallel environments
world = physics.World(num_envs=4096, device="cuda:0")
world.add_robot(robot)

# Synchronous control loop
while world.active:
    state = world.get_state()
    actions = agent.predict(state)
    world.step(actions)
    world.render_debug()

04_HARDWARE_INTEGRATION

Bus_Protocols

CAN-FD8Mbps / 64B Payload
EtherCATSub-100µs cycle time
RS-485Standard legacy support

Safety_Kernel

deterministic safety monitors that override LBM actions if joint torque, velocity, or collision bounds are violated.

Latency
<10µs
Accuracy
99.999%

Need technical assist?

access our private engineering forum or query the optimus agent directly for code generation and training troubleshooting.