BenchmarkPAW-GEN-10-Agentic
Environmente08Place spray bottles upright, nozzle forward (bimanual handover)
- Episodes
- Claude Opus 5.5 30 Claude Fable 5.2 30 DiT-Flow 30 Gemini 4 Argon 30 GPT-6 Astra 30 GR00T N1.7 30 MolmoAct 2 30 Pi 0.5 30 Rho 30
- Total duration
- Claude Opus 5.5 63:30 Claude Fable 5.2 44:41 DiT-Flow 12:35 Gemini 4 Argon 52:16 GPT-6 Astra 101:48 GR00T N1.7 20:27 MolmoAct 2 14:43 Pi 0.5 13:07 Rho 21:59
- Mean duration
- Claude Opus 5.5 127.0 s Claude Fable 5.2 89.4 s DiT-Flow 25.2 s Gemini 4 Argon 104.5 s GPT-6 Astra 203.6 s GR00T N1.7 40.9 s MolmoAct 2 29.4 s Pi 0.5 26.2 s Rho 44.0 s
- Success
- Claude Opus 5.5 97% Claude Fable 5.2 100% DiT-Flow 7% Gemini 4 Argon 83% GPT-6 Astra 87% GR00T N1.7 17% MolmoAct 2 13% Pi 0.5 20% Rho 17%
- Manipulation density
- Claude Opus 5.5 Share of time manipulating, five definitionsJoints moving90.8%End-effector moving80.6%Gripper engaged82.9%Contact force > 5 N79.9%Gripper engaged or in contact90.3% Claude Fable 5.2 Share of time manipulating, five definitionsJoints moving95.0%End-effector moving79.6%Gripper engaged89.4%Contact force > 5 N60.4%Gripper engaged or in contact94.6% DiT-Flow Share of time manipulating, five definitionsJoints moving97.5%End-effector moving95.1%Gripper engaged51.6%Contact force > 5 N58.7%Gripper engaged or in contact74.8% Gemini 4 Argon Share of time manipulating, five definitionsJoints moving87.3%End-effector moving69.3%Gripper engaged96.7%Contact force > 5 N43.2%Gripper engaged or in contact96.7% GPT-6 Astra Share of time manipulating, five definitionsJoints moving78.8%End-effector moving52.4%Gripper engaged65.3%Contact force > 5 N46.8%Gripper engaged or in contact78.3% GR00T N1.7 Share of time manipulating, five definitionsJoints moving100.0%End-effector moving99.3%Gripper engaged70.5%Contact force > 5 N67.6%Gripper engaged or in contact89.1% MolmoAct 2 Share of time manipulating, five definitionsJoints moving100.0%End-effector moving99.4%Gripper engaged66.5%Contact force > 5 N69.1%Gripper engaged or in contact86.2% Pi 0.5 Share of time manipulating, five definitionsJoints moving100.0%End-effector moving98.4%Gripper engaged60.5%Contact force > 5 N67.2%Gripper engaged or in contact84.8% Rho Share of time manipulating, five definitionsJoints moving100.0%End-effector moving99.2%Gripper engaged68.0%Contact force > 5 N63.5%Gripper engaged or in contact89.2%
- SPARC smoothness
- Claude Opus 5.5 Spectral arc length of the motionHow directly the arm moved, read from the shape of its speed spectrum. Unitless measure ≤ -1; the closer to -1, the smoother.The measure ignores how large a movement is, so a large and a small one compare directly. Shown here as the mean over this environment's measured episodes.Cutoff10 HzAmplitude threshold5%Zero-padding16×Signalend-effector translationBalasubramanian et al., On the analysis of movement smoothness, J Neuroeng Rehabil 2015. Claude Fable 5.2 Spectral arc length of the motionHow directly the arm moved, read from the shape of its speed spectrum. Unitless measure ≤ -1; the closer to -1, the smoother.The measure ignores how large a movement is, so a large and a small one compare directly. Shown here as the mean over this environment's measured episodes.Cutoff10 HzAmplitude threshold5%Zero-padding16×Signalend-effector translationBalasubramanian et al., On the analysis of movement smoothness, J Neuroeng Rehabil 2015. DiT-Flow Spectral arc length of the motionHow directly the arm moved, read from the shape of its speed spectrum. Unitless measure ≤ -1; the closer to -1, the smoother.The measure ignores how large a movement is, so a large and a small one compare directly. Shown here as the mean over this environment's measured episodes.Cutoff10 HzAmplitude threshold5%Zero-padding16×Signalend-effector translationBalasubramanian et al., On the analysis of movement smoothness, J Neuroeng Rehabil 2015. Gemini 4 Argon Spectral arc length of the motionHow directly the arm moved, read from the shape of its speed spectrum. Unitless measure ≤ -1; the closer to -1, the smoother.The measure ignores how large a movement is, so a large and a small one compare directly. Shown here as the mean over this environment's measured episodes.Cutoff10 HzAmplitude threshold5%Zero-padding16×Signalend-effector translationBalasubramanian et al., On the analysis of movement smoothness, J Neuroeng Rehabil 2015. GPT-6 Astra Spectral arc length of the motionHow directly the arm moved, read from the shape of its speed spectrum. Unitless measure ≤ -1; the closer to -1, the smoother.The measure ignores how large a movement is, so a large and a small one compare directly. Shown here as the mean over this environment's measured episodes.Cutoff10 HzAmplitude threshold5%Zero-padding16×Signalend-effector translationBalasubramanian et al., On the analysis of movement smoothness, J Neuroeng Rehabil 2015. GR00T N1.7 Spectral arc length of the motionHow directly the arm moved, read from the shape of its speed spectrum. Unitless measure ≤ -1; the closer to -1, the smoother.The measure ignores how large a movement is, so a large and a small one compare directly. Shown here as the mean over this environment's measured episodes.Cutoff10 HzAmplitude threshold5%Zero-padding16×Signalend-effector translationBalasubramanian et al., On the analysis of movement smoothness, J Neuroeng Rehabil 2015. MolmoAct 2 Spectral arc length of the motionHow directly the arm moved, read from the shape of its speed spectrum. Unitless measure ≤ -1; the closer to -1, the smoother.The measure ignores how large a movement is, so a large and a small one compare directly. Shown here as the mean over this environment's measured episodes.Cutoff10 HzAmplitude threshold5%Zero-padding16×Signalend-effector translationBalasubramanian et al., On the analysis of movement smoothness, J Neuroeng Rehabil 2015. Pi 0.5 Spectral arc length of the motionHow directly the arm moved, read from the shape of its speed spectrum. Unitless measure ≤ -1; the closer to -1, the smoother.The measure ignores how large a movement is, so a large and a small one compare directly. Shown here as the mean over this environment's measured episodes.Cutoff10 HzAmplitude threshold5%Zero-padding16×Signalend-effector translationBalasubramanian et al., On the analysis of movement smoothness, J Neuroeng Rehabil 2015. Rho Spectral arc length of the motionHow directly the arm moved, read from the shape of its speed spectrum. Unitless measure ≤ -1; the closer to -1, the smoother.The measure ignores how large a movement is, so a large and a small one compare directly. Shown here as the mean over this environment's measured episodes.Cutoff10 HzAmplitude threshold5%Zero-padding16×Signalend-effector translationBalasubramanian et al., On the analysis of movement smoothness, J Neuroeng Rehabil 2015.
Episode
Held-out benchmark environment
0:002:04