Sim-to-real gap index

Name: Sim-to-real gap index
Creator: sim2real.world
License: https://creativecommons.org/licenses/by/4.0/

What each paper reported in simulation, what it reported on hardware, and the number it did not give. Every value carries a verbatim quote from the paper's own text; an empty cell means the paper does not report it.

paper	task	sim	real	not reported
4D Point Splatting for mmWave Radar Micro-Doppler Differentiable Rendering2611.08342 · hardware not stated · zero-shot	4DPS	not reported	21.9 cmMPJPE on real RT-Pose data for a pose model trained on rendered data alone	rendered-data test MPJPE for the rendered-trained pose model absolute AP for real-trained and rendered-trained HuPR pose models
TactileLab: Sim-to-Real Dexterous Robot Manipulation with Efficient Shear-Sensitive Tactile Simulation2611.07881 · hardware not stated · not stated	IsaacLab	not reported	not reported	success rate in simulation success rate on real robot sim-to-real success-rate gap
An Empirical Study of Reinforcement Learning for Keypoint Positioning of Deformable Linear Objects2611.07261 · hardware not stated · zero-shot	MuJoCo Playground	96.4% for single-point tracking, 98.0% for disturbance recovery, and 88.8% for two-point trackingsimulation success rate	9/10, 10/10, and 8/10 physical trialsreal-world success rate	real-world completion time perception keypoint error sim-to-real success-rate gap
A Practical Causal Framework for Autonomous Mobile Robots: From Simulation to Real-World Human Environments2611.04027 · TIAGo · fine-tuned	PeopleFlow (Gazebo-based social simulator)	not reported	not reported	quantitative simulation task success rate quantitative real-world task success rate quantitative proxemic compliance rate
Coordinate Frame Recovery, Per Class NMS and Null Return Modelling for Sim2Real V2X LiDAR Detection2611.03940 · Ouster OS1-64 · zero-shot	UrbanTwin	0.9251realism score	0.4382detection mAP	synthetic detection mAP final per-class AP on real for vehicle, pedestrian, truck
FlowBench: Counterfactual Imagination Training for Spatial Intelligence2611.03937 · hardware not stated · fine-tuned	AI2-THOR and ProcTHOR	29.5 percentage pointsAccuracy improvement of Qwen3-VL-8B after fine-tuning on FlowBench-Sim-Train, evaluated in simulation	16.9 on real imagesAccuracy improvement of Qwen3-VL-8B after fine-tuning on FlowBench-Sim-Train, evaluated on FlowBench-Real-Test	Absolute post-fine-tune accuracy of Qwen3-VL-8B on FlowBench-Sim-Test Absolute post-fine-tune accuracy of Qwen3-VL-8B on FlowBench-Real-Test Pre-fine-tune accuracy of Qwen3-VL-8B on FlowBench-Real-Test
Geometric Refinement and Distributional Alignment for Sim2Real Roadside LiDAR Detection2611.03936 · hardware not stated · zero-shot	—	0.9126realism score	0.1333detection mAP	detection mAP on synthetic validation data per-class AP for all seven classes on hidden real frames individual Chamfer, MMD, EMD, and FPD scores for the submitted synthetic point clouds
Clinician-AI Collaboration in Endovascular Neurointerventions: In-Vitro User Study of Copilot Assistance Modes for Robotic Navigation Across the Autonomy Spectrum2611.03446 · custom-built endovascular robot · zero-shot	stEVE framework and SOFA	81%success rate	23%success rate+58.0 pp	normalized path ratio for simulation and in-vitro autonomous navigation procedure time for simulation and in-vitro autonomous navigation per-target success rates RICA versus RECA
ScheduleGen: Demonstration Generation via Skill Scheduling for Multi-Arm Imitation Learning2610.39429 · hardware not stated · zero-shot	—	65.5%mean policy success	50–80%success across eight real-world variants	real-world mean success rate per-task real-world success rates number of real-world trials per variant
Explore Where It Matters: Targeted Exploration for Efficient Rover Adaptation2610.38657 · ERNEST rover · online adaptation	DARTS	approximately 90%slope-climbing success	within three to five real-world trialsrecovery of coordinated active-suspension locomotion	real-world slope-climbing success rate sim-to-real performance gap real-world energy or completion time
SimRetract: Zero-Shot Sim-to-Real Robotic Assistance for Retractor Take-Over in Open Surgery2610.38234 · Trossen WidowX AI arm · zero-shot	Isaac Sim	94%grasp success rate on held-out simulated presentations	90%grasp success rate on surgeon-defined real presentations+4.0 pp	real post-release instrument tip offset real axial twist real-world instrument segmentation/plane-estimation error
Omni-IGN: Onboard Image-Goal Navigation for Drones with Omnidirectional Vision2610.36968 · drone with Jetson Orin NX, OAK-4P-New quad-fisheye rig, Livox Mid-360s LiDAR · zero-shot	Habitat, IsaacSim	not reported	not reported	simulation Success Rate (SR) on Habitat benchmarks real-world Success Rate (SR) collision rate in real-world flights
ActiveAlign: Aligning Egocentric Manipulation via Active Perception2610.36859 · Unitree G1 · not stated	MuJoCo	not reported	not reported	manipulation success rate values for Midi Slider/Toggle Switch tasks sim-to-real transfer success rate
From Biological Curves to Soft Modular Robots: Learning-Based Inverse Design with Uncertainty-Guided Refinement2610.36681 · SoftSnap · zero-shot	—	93%shape reconstruction success rate	not reported	physical reconstruction error real-world success rate sim-to-real error

14 rows · extracted by an automated pass, every value quote-verified against the paper text · built 2026-11-07 · gaps.json
This is a record of reporting, not a ranking. A missing number is a finding, not an error: the last column lists metrics a reader needs to judge transfer that the paper does not supply.