benchmark · benchmark--robotwin-2

RoboTwin 2.0

Bimanual data generator and policy-generalization benchmark with public trajectories.

Identity and state

PublicationReleased
DeploymentResearch
Canonical claims16
Last reviewed2026-09-04

Reviewed relationships

StateRelated entityRelationScopeBoundary
No reviewed relations are published for this entity yet.

Canonical evidence claims

Claims are projected from authored tables. A dash remains an explicit missing value; it is never filled from a neighboring product or name match.

MetricValueClass
Asset libraryRoboTwin-OD: 731 objects across 147 categories [benchmarks-r34]vla-manipulation-eval
Claim boundary50 benchmark tasks are distinct from the 10-task code-generation study and 4-task real-world study; those method scores stay in the paper [benchmarks-r34]vla-manipulation-eval
Embodiment in the sourcedual-arm pairings of Aloha-AgileX, ARX-X5, Piper, Franka, and UR5 manipulators [benchmarks-r34]vla-manipulation-eval
Engine / renderervla-manipulation-eval
Evaluation scope5 randomization axes: clutter, lighting, background, tabletop height, and language; clean versus randomized policy evaluation [benchmarks-r34]vla-manipulation-eval
First public datearXiv 2025-06-22 [benchmarks-r34]vla-manipulation-eval
Kindscalable bimanual data generator and policy-generalization benchmark [benchmarks-r34]vla-manipulation-eval
Language-conditionedyes — trajectory-level language is one randomization axis [benchmarks-r34]vla-manipulation-eval
Mediumsimulation benchmark and data generator with a separate sim-to-real study [benchmarks-r34]vla-manipulation-eval
PaperarXiv:2506.18088 [benchmarks-r34]vla-manipulation-eval
Primary metric familytask success under clean and hard randomized conditions on the full 50-task protocol [benchmarks-r34]vla-manipulation-eval
Public codeyes — MIT-licensed official repository [benchmarks-r35]vla-manipulation-eval
Public datayes — object library and pre-collected trajectories are linked from the official project [benchmarks-r35]vla-manipulation-eval
Public leaderboardyes — official leaderboard linked from the project [benchmarks-r35]vla-manipulation-eval
Released trajectories100,000+ pre-collected expert trajectories across 50 tasks and 5 embodiments [benchmarks-r34]vla-manipulation-eval
Task structure50 dual-arm tasks in the paper; the project describes over 50 supported tasks [benchmarks-r34] [benchmarks-r35]vla-manipulation-eval

Source ledger

What is not inferred

No fuzzy entity matching, adjacent-column borrowing, universal compatibility, commercial maturity, or checkpoint-to-product identity. Absence of a relation means unreviewed or unsupported here—not proven incompatibility.