Benchmark Guides

This section provides end-to-end evaluation workflows organized by benchmark. Each guide covers environment setup, example configs, step-by-step commands, and advanced usage.

Benchmark	Description	Guide
RealWorld	Franka real-robot evaluation and deployment	Real-World Evaluation
BEHAVIOR-1K	Large-scale household scene simulation	BEHAVIOR-1K Evaluation
LIBERO	Robotic manipulation benchmark with Spatial / Object / Goal / Long / 90 suites	LIBERO Evaluation
ManiSkill OOD	ManiSkill out-of-distribution generalization evaluation	ManiSkill OOD Evaluation
PolaRiS	Tabletop manipulation simulation platform	PolaRiS Evaluation
RoboTwin	Bimanual manipulation simulation with multiple tasks	RoboTwin Evaluation
SIMPLE	Whole-body humanoid evaluation with Psi0	SIMPLE Evaluation

Note

Benchmarks such as IsaacLab and MetaWorld do not yet have example configs under evaluations/. For evaluation, refer to the corresponding training docs in RL with Embodied Simulators and use the config fallback mechanism with YAMLs under examples/embodiment/config/.