Robohouse ’26 Library
Contents

Appendix B

What Could Not Be Verified

about 1 minutes

  • Quantitative results from Ajoudani, Tsagarakis & Bicchi (2012), IJRR — paywalled; do not cite 2012 numbers.
  • Smith, Hargrove, Lock & Kuiken (2011) optimal window length — commonly quoted as 150–250 ms, unverified.
  • Young et al., Gao et al. and Côté-Allard et al. electrode-shift figures — cited second-hand via Tanaka 2025.
  • Hahne et al. (2014) per-method regression R2R^2 values.
  • Zhuang et al. (2019), Nat. Mach. Intell. — participant counts and success percentages paywalled.
  • ForceBand dataset — announced, not released; subject count (4) from limitations section.
  • DexEMG — no data release, no control rate, no latency reported.
  • Open X-Embodiment force/torque fraction — no defensible number exists; do not cite a percentage.
  • ABB × PSYONIC collaboration — press release only, no hours, demos or benchmark.
  • Gemini Robotics On-Device 2 "<200 examples" — model card only, no technical report.
  • "Manus Robotics Hemyo" — no evidence this product exists; Manus uses optics-based, not electromyographic, sensing.