Summary
The first benchmark targeting humanoid dexterous manipulation specifically in organic-chemistry lab settings. It reconstructs 30+ functionally faithful lab-instrument assets via real-to-sim modeling of an actual chemistry workstation, and models a “Tianyi 2.0” humanoid performing precision-critical operations (e.g. solid-solid powder transfer) with articulated instruments, particle-based powder physics, and closed-loop instrument state readouts.
Key Contributions
- The first simulation benchmark of its kind for lab-science humanoid manipulation.
- A real-to-sim asset library of 30+ chemistry-lab instruments.
- Particle-based powder physics combined with instrument-readout coupling for closed-loop task verification.
Strengths
- A novel, under-served application domain (scientific lab automation) with concrete simulation infrastructure.
- Addresses genuinely hard contact dynamics (granular/powder physics) rarely modeled in manipulation benchmarks.
Weaknesses
- A single-domain (organic chemistry) benchmark, with unclear transfer to broader dexterous-manipulation research.
- No baseline policy performance numbers were found in available sources.
Open Questions
- What policies were evaluated, and what are baseline success rates?
- Is there sim-to-real validation on physical Tianyi 2.0 hardware, or is this sim-only?
Significance
Opens a genuinely new application niche — scientific laboratory automation — for humanoid dexterous manipulation research, an area largely absent from prior benchmarks in this vault.