Simulated Robot Action Chunking on Kinetix (full-data)
83.8Overall ReturnTrain Hard RTC
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Train Hard RTCRTC Stage=Training-time, RTC Variant=Hard2026.05 | 83.8 | 81.5 | 73.2 | 2.416 | 1.07 | 1.01 | |
| Train Soft RTC (N=1)RTC Stage=Training-time, RTC Variant=Soft, Window Size (N)=12026.05 | 83.2 | 80.9 | 72.4 | 2.413 | 1.04 | 0.95 | |
| Train Soft RTC (N=2)RTC Stage=Training-time, RTC Variant=Soft, Window Size (N)=22026.05 | 79.2 | 78.1 | 70.2 | 2.185 | 1.05 | 0.95 | |
| Inference Hard RTCRTC Stage=Inference-time, RTC Variant=Hard2026.05 | 77.7 | 78.3 | 69.3 | 2.501 | 2.63 | 2.19 | |
| Inference Soft RTCRTC Stage=Inference-time, RTC Variant=Soft2026.05 | 77.3 | 77.8 | 67.8 | 2.471 | 2.68 | 2.21 | |
| Base NaiveRTC Variant=Naive2026.05 | 66.2 | 69.1 | 53.1 | 2.533 | 1 | 1 |