Evidence Summary Table

Claim	Value	Source	Grade
Electromechanical delay (voluntary)	49.73 ± 6.99 ms	Begovic et al. 2014, Front. Physiol.	A
Optimal myoelectric controller delay	100 ms (fast) / 125 ms (slow)	Farrell & Weir 2007, IEEE TNSRE	A
Electrode shift, 2 cm	−7.6% to −20%	Tanaka et al. 2025, Sensors (consolidating)	A/B
Re-donning shift magnitude	1.16 ± 0.34 cm	Hwang et al. 2017, PLOS ONE	A
Limb-position effect	3.8% → 18% error	Fougner et al. 2011, IEEE TNSRE	A
Cross-session degradation	~10–12 pp	Rehman et al. 2018, Sensors	A
Skin impedance stabilisation	~15 min	Sousa et al. 2023, Sensors	A
Cross-user hand-pose error (best)	12.2°	emg2pose, NeurIPS 2024	A
Generic sEMG, no calibration	>90% held-out-user; 11,236 participants	Kaifosh et al. 2025, Nature 645	A
EMG vs conventional interfaces	26% slower, 58× error, 2.8× TLX	M4Bench 2025, Front. Robot. AI	A
Tele-impedance vs constant stiffness	83.3% vs 78.3% / 60.0%	Fani et al. 2018, IEEE RAM	A
Tele-impedance under 1 s delay	12.33 N vs 24.31 N peak force	Laghi et al. 2020, IJRR	A
Vision-derived stiffness (no EMG)	0.87 vs 0.76 / 0.24 success; TLX 38.7 vs 49.3 / 52.4	Stiffness Copilot 2026	C
EMG force annotation → policy	87% pick–squeeze–place; 0/10 baseline squeeze	ForceBand 2026	C
Force relabelling of actions	near-zero → 76%	DexForce, RA-L	A
Force as raw observation	0.475–0.575 vs 0.500 baseline (≈ nil)	FoAR, RA-L	A
Gated force fusion	0.875 vs 0.500	FoAR, RA-L	A
Tactile can hurt	0.95 → 0.80 (Daimon, pick-and-place)	TaCo 2026	C
Hallucinated tactile ≈ real	45% vs 35% (cup nesting)	FELT 2026	C
VLA new-task adaptation	10–150 demos	OpenVLA, CoRL 2024	A
LoRA vs full fine-tune	68.2% vs 69.7%; 1.4% params; 8× compute	OpenVLA, CoRL 2024	A
VLA new-embodiment adaptation	<200 examples	Gemini Robotics On-Device 2	D
Curation with influence functions	SOTA at <33% of data	CUPID, CoRL 2025	A
Curation with smoothness (SPARC)	+16% at 1/6 data	RINSE 2026	C
Mixed-operator penalty	72% → 42% (Transport)	robomimic, CoRL 2021	A
Policy noise at low data	69.7% vs 48.3% (50 eps)	Belkhale et al., NeurIPS 2023	A
Prosthesis as data-collection hand	79.3% over 450 trials	HannesImitation 2025	C
Shared autonomy, HD-EMG, in home	738 s → 517 ± 62 s (−30.0%)	Yang et al. 2026	C
Shared autonomy cognitive load	0.62 s vs 0.74 s DRT (−24%)	George lab, Nat. Commun. 2025	A