Overview

irsim-x is a C++-accelerated fork of IR-SIM, an open-source, Python-based lightweight robot simulator for navigation, control, and reinforcement learning. The core simulation loop (LiDAR raycasting, collision detection, kinematics stepping, and A* planning) has been ported to C++ via pybind11, with AVX2 SIMD acceleration for LiDAR and batch multi-environment simulation.

The original Python-only paths have been removed — C++ is the only execution path. Compared to the original IR-SIM, IR-SIM-X achieves 5–50× faster LiDAR raycasting, 8× faster end-to-end step on sparse environments, and supports batch SIMD simulation across up to 1024 parallel environments.

Key Features

• C++ LiDAR — AVX2 SIMD circle/rect raycast + spatial hash grid for polygon/linestring, 5–50× faster
• C++ Collision Detection — SAT (Separating Axis Theorem) with ear-clip triangulation for concave polygons
• C++ Kinematics — diff, omni, omni_angular, ackermann steering
• C++ SimWorld — unified step, collision, dynamic obstacle (circle/rect/polygon/linestring) management
• Batch SIMD — up to 1024 environments with cross-environment SIMD LiDAR and kinematics
• FMCW LiDAR — C++ accelerated per-beam radial velocity computation
• YAML-driven — same simple configuration as IR-SIM, no code changes needed for existing scenarios

Installation

Requires Python >= 3.10 and a C++17 compiler (for building the extension)

pip install irsim-x

# With keyboard control and extras
pip install irsim-x[all]

From source

git clone https://github.com/Ethereal1024/ir-simX.git
cd ir-simX
pip install -e .

Quick Start

A minimal example — a differential-drive robot navigates toward a goal:

import irsim

env = irsim.make('robot_world.yaml')

for i in range(300):
    env.step()
    env.render()
    if env.done():
        break

env.end()

YAML Configuration (robot_world.yaml):

world:
  height: 10
  width: 10
  step_time: 0.1

robot:
  kinematics: {name: 'diff'}
  shape: {name: 'circle', radius: 0.2}
  state: [1, 1, 0]
  goal: [9, 9, 0]
  behavior: {name: 'dash'}

For more examples, see the usage directory.

Support

Performance

Projects Using IR-SIM

Academic Publications

• [RAL & ICRA 2023] rl-rvo-nav — Reinforcement learning-based RVO behavior for multi-robot navigation.
• [RAL & IROS 2023] RDA_planner — Accelerated collision-free motion planner for cluttered environments.
• [T-RO 2025] NeuPAN — Direct point robot navigation with end-to-end model-based learning.

Community Projects

• DRL-robot-navigation-IR-SIM — Deep reinforcement learning for robot navigation.
• AutoNavRL — Autonomous navigation using reinforcement learning.
• IRSIM-3DGS-Bridge — A closed-loop bridge from 3D Gaussian Splatting scenes to IR-SIM planning/following.

Citation

IR-SIM-X is a C++-accelerated fork of IR-SIM. If you find it useful, please consider citing the original IR-SIM paper:

@article{han2026ir,
  title={IR-SIM: A Lightweight Skill-Native Simulator for Navigation, Learning, and Benchmarking},
  author={Han, Ruihua and Wang, Shuai and Li, Chengyang and Gao, Rui and Wang, Xinyi and Liu, Zhe and Li, Guoliang and Lu, Yupu and Hao, Qi and Pan, Jia and Zhao, Hengshuang},
  journal={arXiv preprint arXiv:2606.08729},
  year={2026}
}

License

irsim-x is released under the MIT License. The original IR-SIM is also MIT-licensed — see the original repository for details.