SPEAR: Realistic Embodied AI Simulator

adobe-research · hf · 2026-07-17

SPEAR is a simulator for realistic embodied AI research, functioning primarily as a Python library that connects to and programmatically controls any Unreal Engine application via a modular plugin architecture.

The paper highlights several key improvements:

The authors also propose a high-level programming model that organizes UE workloads into graphs with arbitrary data dependencies, executing deterministically within a single UE frame. The paper showcases various use cases: controlling multiple embodied agents with different action spaces, generating city-scale realistic scenes, manipulating procedural content generation systems, synchronizing multi-view face renders, conducting interactive co-simulation with MuJoCo, and editing scenes via natural language using AI coding assistants.

Original post →

More from Embodied

Embodied channel →