terasim 0.1.0

Generative AV testing environment for unknown unsafe events discovery

Generative Autonomous Vehicle Testing Environment for Unknown Unsafe Events Discovery

---

Website • Paper • Examples • Community

About

TeraSim is an open-source traffic simulation platform designed for naturalistic and adversarial testing of autonomous vehicles (AVs). It enables high-speed, AI-driven testing environment generation to expose AVs to both routine and rare, high-risk driving conditions.

Developed with researchers, AV developers, and regulators in mind, TeraSim is designed to better support ISO 21448 (SOTIF) and ISO 34502 compliance, providing a scalable, automated, and unbiased AV evaluation framework.

Built upon the open-source traffic simulation software SUMO (Simulation of Urban MObility), TeraSim extends its capabilities to provide specialized features for autonomous vehicle testing.

🎥 Demo Video

TeraSim is built upon a series of foundational academic works in autonomous vehicle testing:

• NDE (Paper | Code): Learning naturalistic driving environment with statistical realism. Yan, X., Zou, Z., Feng, S., et al. Nature Communications 14, 2037 (2023).

• NADE (Paper | Code): Intelligent driving intelligence test for autonomous vehicles with naturalistic and adversarial environment. Feng, S., Yan, X., Sun, H. et al. Nature Communications 12, 748 (2021).

• D2RL (Paper | Code): Dense reinforcement learning for safety validation of autonomous vehicles. Feng, S., Sun, H., Yan, X., et al. Nature 615, 620–627 (2023).

---

🌟 Key Features

✅ Generative Driving Environment Testing
→ Adaptive and interactive environments replace static, manually designed scenarios.
→ Automatically uncovers unknown unsafe events, enhancing AV safety validation.
→ Scalable and efficient, reducing manual effort while expanding test coverage.

✅ Naturalistic & Adversarial Driving Environments (NADE)
→ Real-world traffic behavior modeling based on large-scale naturalistic driving data.
→ Injects corner cases (e.g., jaywalking pedestrians, sudden lane changes) to rigorously test AV safety.

✅ Scalable & Automated AV Testing
→ AI-driven naturalistic and adversarial driving environment accelerates AV validation by 1,000x - 100,000x compared to real-world testing.
→ Dynamically adapts test cases to urban, highway, and mixed-traffic conditions.

✅ Seamless Integration with Third-Party Simulators
→ Works with CARLA, Autoware, and more.
→ API-driven design enables plug-and-play simulation for integration with third-party simulators.

✅ City-Scale AV Testing with TeraSim-Macro
→ Extends simulations from single intersections to entire cities, supporting policy-level AV impact analysis.

✅ Multimodal Inputs & AI-Assisted Environment Creation
→ TeraSim-GPT enables language-driven environment customization.
→ Define test cases in natural language: "Create a left-turn driving environment at a busy intersection."

---

🛠️ System Architecture

TeraSim is modular, allowing users to customize and extend simulations easily.

📌 Core Components:

• TeraSim: Base simulation engine for generating AV test environments.
• TeraSim-NDE-NADE: Realistic & adversarial driving environments for safety evaluation.
  • Vehicle Adversities (e.g., aggressive cut-ins, emergency braking).
  • VRU Adversities (e.g., jaywalking pedestrians, erratic cyclists).
• TeraSim-Service: RESTful API service built with FastAPI for seamless integration with popular simulators like CARLA and AWSim. Enables standardized communication and control.
• TeraSim-Macro (coming soon): Enables mesoscopic city-scale AV testing.
• TeraSim-Data-Zoo (coming soon): Repository for real-world driving data (Waymo, NuScenes, NuPlan).
• TeraSim-GPT (coming soon): AI-powered multimodal user input handling for environment customization.

📌 Plug-and-Play Compatibility:
✅ SUMO-based microsimulation
✅ CARLA & Autoware integration
✅ Real-world dataset support

---

🔧 Installation

Currently, TeraSim is under active development. Please install it from source using poetry (required) and Anaconda (optional).

conda create -n terasim python=3.10
conda activate terasim

git clone https://github.com/mcity/TeraSim.git
cd TeraSim
poetry install

🚀 Why TeraSim?

🔍 Uncover Hidden AV Risks
→ Dynamically generates realistic and adversarial traffic environments, identifying corner cases.

⚡ Automated & Scalable
→ Uses AI to generate simulations across cities, with 1000x faster testing efficiency than real-world methods.

🔗 Seamless Integration
→ Plugin-based design works with existing AV stacks & third-party simulators.

📢 Open-Source & Extensible
→ Encourages industry collaboration for safer, more reliable AV deployment.

---