List your company

Robotic Manipulation on RLBench (test)

90.5Average Success RateTGM-VLA
-2.26821.81645.969.984Feb 16, 2024Jun 18, 2024Oct 20, 2024Feb 21, 2025Jun 25, 2025Oct 27, 2025Feb 28, 2026

Evaluation results

MethodLinks
TGM-VLAReported Hardware=2A6000×1d2026.0290.51009988989310099100969489100708783948060———————————————————————————————————————
TGM-VLA w/o ITM2026.0288.8100999299969898100959286100658282917850———————————————————————————————————————
BridgeVLAReported Hardware=48H100×20h2026.0288.298.49687.210092.899.210010099.28887.210076.881.673.68860.858.4———————————————————————————————————————
TGM-VLA w/o CTM2026.0288.210097989696959998949488100627980927347———————————————————————————————————————
SAM2ACTReported Hardware=8H100×12h or 2A6000×5d2026.0286.81008699989699999998938983767875846447———————————————————————————————————————
FlowRAMNumber of demonstrations=1002025.0684.9100100929410092961009696849277.36186724842———————————————————————————————————————
RVT-2Number of demonstrations=1002025.0681.4100921009999961009996958874806966403538———————————————————————————————————————
RVT-2Train time=0.83 days, Inf. Speed=20.6 fps, Resolution=128x128, Number of demonstrations=1002024.0681.41009210099999610099969588748069664035381.5——————————————————————————————————————
RVT-2Reported Hardware=8V100×20h2026.0281.4100921009999961009996958874806966403538———————————————————————————————————————
3D Diffuser ActorNumber of demonstrations=1002025.0681.398.497.68496.899.2969610097.693.682.489.668.347.285.665.64424———————————————————————————————————————
3D Diffuser ActorView setup=multi-view2024.0281.398.497.68496.899.2969610097.693.682.489.668.347.285.665.64424———————————————————————————————————————
MLA2025.0981——100——————75———————————90——————60——————————————55———95—95—10050——900.03
3D Diffuser ActorAgent Type=3D next-best-pose, Priors=depth, keypoints, scene bounds, motion-planners, Training Demos=50, Protocol=multi-task2024.0773.846.799.3——99.3——————71.3———————10029.376876.793.35882.796.79492.79283.336.777.3966610072.7944—————————————————
HybridVLA2025.0966——80——————70———————————30——————50——————————————25———85—95—10050——750.05
Act3DTrain time=5.0 days, Resolution=256x256, Number of demonstrations=1002024.06659993929494909292958047931295127832.8——————————————————————————————————————
π02025.0965——85——————65———————————90——————5——————————————25———85—100—9010——950.04
DreamVLA*status=re-implemented2025.0965——25——————85———————————50——————35——————————————35———95—100—9555——750.05
FDPPolicy=FDP2025.1263.9———————————————————————————————————————————————7210075.59327.578614.5——
Act3DView setup=multi-view2024.0263.293.69686.495.294.491.296.880.895.259.232.878.449.667.22429.63.2———————————————————————————————————————
RVTTrain time=1.0 days, Inf. Speed=11.6 fps, Resolution=128x128, Number of demonstrations=1002024.0662.910081.6728893.6885299.291.2914871.228.826.449.611.23642.8——————————————————————————————————————
RVTView setup=multi-view2024.0262.910081.6728893.6885299.291.2914871.228.826.449.611.2364———————————————————————————————————————
RVTReported Hardware=8V100×1d2026.0262.910081.6728893.6885299.291.2914871.228.826.449.611.2364———————————————————————————————————————
GENIMAAgent Type=RGB-to-joint, Training Demos=50, Protocol=multi-task2024.0749.676.712.7——71.3——————77.3———————50269479.385.348.746.74669.318.7369072.714.712.726.711.34821.393.312—————————————————
PerActNumber of demonstrations=1002025.0649.492.8745270.48851.255.289.68644.817.68826.42.4285.616.82.4———————————————————————————————————————
PerActTrain time=16.0 days, Inf. Speed=4.9 fps, Resolution=128x128, Number of demonstrations=1002024.0649.492.8745270.48851.255.289.68444.817.68826.42.4285.616.82.44.4——————————————————————————————————————
PerActView setup=multi-view2024.0249.492.8745270.48851.255.289.68644.817.68826.42.4285.616.82.4———————————————————————————————————————
PerActReported Hardware=8V100×16d2026.0249.492.8745270.48851.255.289.68444.817.68826.42.4285.616.82.4———————————————————————————————————————
PolarNetResolution=128x128, Number of demonstrations=1002024.0646.49656521008032369284404484481241204.8——————————————————————————————————————
PolarNetView setup=multi-view2024.0246965652100803236928440448448124120———————————————————————————————————————
SpatialVLA2025.0946——20——————15———————————40——————15——————————————30———80—80—8525——700.03
DPPolicy=DP2025.1245.6———————————————————————————————————————————————33.51007765.564193.52——
HiveFormerResolution=128x128, Number of demonstrations=1002024.0645.384642810080685276768085280320805.2——————————————————————————————————————
HiveFormerView setup=multi-view2024.02458464281008068527676808528032080———————————————————————————————————————
MoDEPolicy=MoDE2025.1245———————————————————————————————————————————————319770.550.56010.5373.5——
GEM3DBackbone=PerceiverIO, Pre-training=100K iterations, Policy Training=12K iterations, SE(3) augmentations=true2026.0144.2———44.354—44.7—29.711.7—65.77.7—————1—————————67.3——————————————32.7406172——————————
SDPPolicy=SDP2025.1242.1———————————————————————————————————————————————241006842.558.56.5361——
UP-VLA2025.0942——10——————20———————————55——————15——————————————20———80—80—6535——400.04
OpenVLA2025.0940——55——————20———————————15——————20——————————————5———60—85—7530——350.02
ACTAgent Type=RGB-to-joint, Training Demos=50, Protocol=multi-task2024.0739.631.322——59.3——————82.7———————32.7184269.375.34022.622.6813.343.399.365.34414.715.322.744.737.345.319.3—————————————————
ManiGaussian (w Distill)Backbone=PerceiverIO, Distillation=true2026.0133.3———35.342.7—31—20.35.3—54.35—————3—————————64.3——————————————16.333.748.743——————————
ManiGaussianBackbone=PerceiverIO, Representation=3D Gaussian Splatting2026.0131.5———3044—27.3—5.78—57.32—————3.08—————————66.3——————————————19.338.330.749.3——————————
GNFactorBackbone=PerceiverIO, Representation=NeRF-based2026.0129.2———3240.7—32—115.7—51.73.7—————3.33—————————62——————————————2235.73519——————————
PerActBackbone=PerceiverIO, Representation=Voxel-based 3D2026.0121.9———28.343.7—19—10.73—38.74—————4.58—————————51.3——————————————8.326.71613.7——————————
SuSIEAgent Type=RGB-to-joint, Training Demos=50, Protocol=multi-task2024.0721.87.30——32.7——————67.3———————5.3021.33626.62610224.6124.64074.7146.71724506—————————————————
C2F-ARM-BCResolution=128x128, Number of demonstrations=1002024.0620.1721602068424241288200004805.8——————————————————————————————————————
C2F-ARM-BCView setup=multi-view2024.0220.172160206842424128820000480———————————————————————————————————————
DiffusionPolicyAgent Type=RGB-to-joint, Training Demos=50, Protocol=multi-task2024.072.32.70——20.7——————0———————0003.36.7002020.71.322.702001.3024—————————————————
Image-BC (CNN)Resolution=128x128, Number of demonstrations=1002024.061.30000880040040000007.4——————————————————————————————————————
Image-BC (ViT)Resolution=128x128, Number of demonstrations=1002024.061.300001600000000000007.7——————————————————————————————————————
Claude-sonnet-4-52026.02—51————————————63.2——————————————————————————3036.720——————————————
DeepSeek-V32026.02—70————————————64.3——————————————————————————904232.5——————————————
Gemini-2.5-pro2026.02—85.6————————————51.5——————————————————————————9546.730——————————————
GPT-52026.02—58.2————————————45——————————————————————————8033.315——————————————
GSRModel size=8B2026.02—82————————————68.5——————————————————————————92.552.541.2——————————————
PERACTNumber of demonstrations=10, Number of evaluation episodes=25, Random seeds=3, Input modality=RGB-D2024.02—20327272323612—681628——56160—4———————————————————————————————————————
Qwen3Model size=8B2026.02—51————————————56.4——————————————————————————2533.326.7——————————————
RVTNumber of demonstrations=10, Number of evaluation episodes=25, Random seeds=3, Input modality=RGB-D2024.02—45.3354.67769622.6797.335.332.6769.3391.3328—065.3338.675.33—2.67———————————————————————————————————————
VPDDNumber of demonstrations=10, Number of evaluation episodes=25, Random seeds=3, Input modality=RGB2024.02—6070.674088.6737.3366.675673.3373.332461.33—5658.6770.6730.67—30.67———————————————————————————————————————