TeraSim: A City-Scale Agentic World Model Platform for Physical AI

Role: Core Developer | Inchor (SaferDrive AI) Period: Dec 2025 – Present

Overview

TeraSim is a city-scale simulation platform serving as the foundational infrastructure for agentic world models under the Physical AI framework, enabling naturalistic and adversarial testing of autonomous vehicles at urban scale.

Core Module — CNDE (Conditional Neural-network Driving Environment)

• CNDE serves as the behavioral-layer foundation model within the TeraSim ecosystem, integrating MotionTransformer for neural network-based driving behavior generation.
• Refactored the CNDE architecture from a dual-state management system to a stateless prediction pipeline, where TeraSim manages all vehicle states and CNDE operates as a pure prediction function.
• Implemented vehicle state synchronization, history tensor construction, and coordinate transformation for large-scale MotionTransformer inference.

Milestones — NVIDIA & City of San Jose Joint Project

• Led a joint project with NVIDIA and the City of San Jose on agent-driven Smart City operation, building an integrated “City Operations Agent” and digital twin that can perceive, simulate, and optimize traffic operations.
• Designed and implemented the full simulation pipeline including city-scale vehicle injection, dynamic routing, and multi-phase demo orchestration.
• Mar 2026: Active deployment with NVIDIA and the City of San Jose as an Agentic Incident-to-Operation pipeline, bringing TeraSim from simulation into real-world city operations.
• Jun 2026: First draft release of Agent-Sim, utilizing LLMs and VLMs for safety-critical scenario generation.

Tech Stack: Python, SUMO, PyTorch, CUDA, Docker, FastAPI