A Survey of Robotic Navigation and Manipulation with Physics Simulators in the Era of Embodied AI

A comprehensive survey examines how physics simulators address the sim-to-real gap in embodied AI, focusing on navigation and manipulation tasks. The research provides benchmarks, metrics, and platform comparisons to help developers select appropriate simulation tools while accounting for hardware constraints.

This survey addresses a critical bottleneck in embodied AI development: the gap between training agents in simulation and deploying them in physical environments. Training robots directly in the real world remains prohibitively expensive, time-consuming, and dangerous, making simulation-based training essential. However, the persistent sim-to-real gap—where policies learned in simulation fail in real-world conditions—limits practical applications. By systematically analyzing physics simulators' properties, features, and hardware requirements, this work advances the field beyond previous surveys that overlooked crucial simulation characteristics affecting transfer success.

The research emerges from growing investment in embodied AI across robotics, autonomous systems, and AI development. Major tech companies and startups increasingly prioritize sim-to-real transfer as the pathway to scalable robot training. The survey's focus on benchmark datasets and metrics directly supports this trend by providing standardized evaluation frameworks that have been lacking in the field.

For developers and researchers, this resource significantly reduces selection friction when choosing simulation platforms. Access to comprehensive comparisons of navigation and manipulation features, hardware requirements, and established benchmarks accelerates development cycles and improves reproducibility across teams. This standardization effect strengthens the entire embodied AI ecosystem by enabling better knowledge transfer and collaborative progress.

Looking ahead, the consolidation of simulation standards may drive investment toward simulation platforms that meet these emerging benchmarks. Researchers should monitor whether specific simulators gain dominance based on these comparative analyses, as market consolidation in the simulation layer could create bottlenecks or opportunities for platform providers serving the expanding embodied AI sector.

• →Physics simulators remain essential for embodied AI training despite persistent sim-to-real transfer challenges.
• →Comprehensive benchmarking of simulation platforms enables researchers to match tools to specific navigation and manipulation tasks.
• →Hardware constraints significantly influence simulator selection, requiring careful consideration beyond software features alone.
• →Standardized metrics and datasets improve reproducibility and accelerate progress in embodied AI research.
• →The survey establishes a resource framework that reduces friction in selecting simulation platforms for robot training.