Nano-sim

Unity-based autonomous-driving simulator · Individual · Jun 2025 — Jun 2026

#Unity#Sim2Real#NeuralRendering#1/10Scale

NANO-SIM

Autonomous Driving Simulator

What it is

Nano-sim is an autonomous-driving simulator for validating models and algorithms, generating data, and closed-loop training. It targets 1/10-scale cars, inspired by platforms like F1TENTH. Students enter autonomous-driving competitions but have no simulator that reproduces the competition environment, which makes validating a model — and trusting its results — difficult. Nano-sim is a personal project built to close that gap.

The International Robot Contest track, in reality — The International Robot Contest arena (left) recreated one-to-one as a digital twin in Unity (right).

The same track recreated in Unity — The International Robot Contest arena (left) recreated one-to-one as a digital twin in Unity (right).

Features

Vehicle dynamicsImplemented as a kinematic model in Unity. At 1/10 scale the car's mass and speed are too small for full vehicle dynamics to matter, so a kinematic model captures its motion faithfully while staying stable and cheap to simulate.

Scene & assetsThe vehicle is modeled on the platform I actually run, and the scene reproduces the competition environment hosted by IROC. Barriers, traffic lights, and other course equipment are included and can be placed freely, so you can design your own scenarios.

Sensor pipelinesCamera, LiDAR, GNSS, and IMU are all implemented. Camera and LiDAR support configurable noise, and their intrinsic and extrinsic parameters can be adjusted.

How I used it

I train reinforcement-learning driving policies inside Nano-sim and transfer them to a physical 1/10-scale car. The same approach produced the lane-following agent behind the International Robot Contest simulator track, where a policy trained in Unity transferred to the real vehicle through Gaussian noise injection.

Read the docs Source on GitHub