Spaces:
Running
Running
| <html lang="en"><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>DVIDIA 路 PiPER contact benchmark</title><style>body{font:18px/1.6 Georgia,serif;margin:auto;max-width:1000px;padding:24px;background:#fafaf7;color:#292d27}a{color:#396751}.scroll{overflow:auto}td,th{padding:10px;text-align:left;border-bottom:1px solid #ddd}table{border-collapse:collapse;font:14px/1.5 system-ui}code{overflow-wrap:anywhere}</style><main><p>DVIDIA RESEARCH</p><h1>PiPER: grasp, carry, release</h1><p>Authored PiPER free-block contact task, synthetic scenes and privileged native state. No learned policy, shoe competence, calibrated contact or physical robot qualification.</p><p>9 planned attempts 路 confirmation cohort. Every failed or invalid attempt stays in the denominator.</p><p><a href="protocol.json">Frozen protocol</a> 路 <a href="report.json">Full report</a> 路 <a href="integrity.json">Integrity</a> 路 <a href="licenses/agilex_piper/LICENSE">Original MIT asset license</a></p><ul><li>nominal / placement: 2/2 completed; 2 valid.</li><li>nominal / replay_open_jaw: 0/2 completed; 2 valid.</li><li>nominal / hold: 0/2 completed; 2 valid.</li><li>low_friction / placement: 0/1 completed; 1 valid.</li><li>low_friction / replay_open_jaw: 0/1 completed; 1 valid.</li><li>low_friction / hold: 0/1 completed; 1 valid.</li></ul><div class="scroll"><table><thead><tr><th>Scene</th><th>Control</th><th>Outcome</th><th>Reason</th><th>Wall seconds</th><th>Evidence</th></tr></thead><tbody><tr><td>nominal-00</td><td>placement</td><td>Completed</td><td>success</td><td>2.285</td><td><a href="attempts/nominal-00--placement.json">Complete attempt</a></td></tr><tr><td>nominal-00</td><td>replay_open_jaw</td><td>Failed</td><td>timeout</td><td>1.45</td><td><a href="attempts/nominal-00--replay_open_jaw.json">Complete attempt</a></td></tr><tr><td>nominal-00</td><td>hold</td><td>Failed</td><td>timeout</td><td>1.386</td><td><a href="attempts/nominal-00--hold.json">Complete attempt</a></td></tr><tr><td>nominal-01</td><td>placement</td><td>Completed</td><td>success</td><td>1.627</td><td><a href="attempts/nominal-01--placement.json">Complete attempt</a></td></tr><tr><td>nominal-01</td><td>replay_open_jaw</td><td>Failed</td><td>timeout</td><td>1.438</td><td><a href="attempts/nominal-01--replay_open_jaw.json">Complete attempt</a></td></tr><tr><td>nominal-01</td><td>hold</td><td>Failed</td><td>timeout</td><td>1.43</td><td><a href="attempts/nominal-01--hold.json">Complete attempt</a></td></tr><tr><td>low_friction-00</td><td>placement</td><td>Failed</td><td>timeout</td><td>1.67</td><td><a href="attempts/low_friction-00--placement.json">Complete attempt</a></td></tr><tr><td>low_friction-00</td><td>replay_open_jaw</td><td>Failed</td><td>timeout</td><td>1.489</td><td><a href="attempts/low_friction-00--replay_open_jaw.json">Complete attempt</a></td></tr><tr><td>low_friction-00</td><td>hold</td><td>Failed</td><td>timeout</td><td>1.436</td><td><a href="attempts/low_friction-00--hold.json">Complete attempt</a></td></tr></tbody></table></div><h2>Measurement scope</h2><p>Attempt wall includes asset verification, compilation, settling, authored planning/IK, physics, contact scoring and recorded state construction. Protocol/report/trace disk serialization, rendering, training, physical sensing, transport and actuation are excluded. Control computation windows are not sensor-to-actuator reaction times.</p><p>Six arm commands use radians; the one jaw command is single-finger travel in metres. A displayed block task does not establish shoe manipulation or hardware transfer.</p><p><code>{"platform": "macOS-26.5.2-arm64-arm-64bit", "machine": "arm64", "python": "3.12.9", "logical_cpu_count": 10, "gpu_used": false, "episode_runtimes": [{"system": "Darwin", "machine": "arm64", "python": "3.12.9", "mujoco_version": "3.15.0", "numpy_version": "2.5.3"}], "process_lifetime_peak_rss_bytes": 759169024, "memory_scope": "Process lifetime high-water RSS before final inspection/export, not per-episode memory or minimum RAM; null unavailable."}</code></p><footer><a href="https://research.dvidia.org/">DVIDIA research</a> 路 <a href="https://github.com/Dvidia-Inference/dvidia-training">GitHub</a> 路 <a href="https://x.com/iammrriver">@iammrriver</a> 路 <a href="mailto:hello@dvidia.org">hello@dvidia.org</a></footer></main></html> |