89.4%
▲ 40 pts since 2022
average RLBench success rate of EquAct, the best method in 2025
20222025
What the chart shows
RLBench tests robot learning methods on simulated manipulation tasks such as picking up objects, stacking items and operating simple mechanisms, here 18 tasks with 100 demonstrations each. The best average success rate rose from 49.4% (PerAct) in 2022 to 65% (Act3D) in 2023, 86% (ARP+) in 2024 and 89.4% (EquAct) in 2025.
The AI Index cautions that these are relatively short tasks in a controlled simulation, far from household work (see the related chart).
Key numbers
| Measure | Value |
|---|---|
| 2022: PerAct | 49.4% |
| 2023: Act3D | 65.0% |
| 2024: ARP+ | 86.0% |
| 2025: EquAct | 89.4% |
| Change, 2022 to 2025 | +40.0 percentage points |
How to read it
- Results as of January 2026, per the AI Index.
- The report text names SAM2Act (86.8%) as the leader before EquAct; the data lists ARP+ at 86% for 2024. We use the data.
Source: AI Index, 2025, via the AI Index Report 2026 (Stanford HAI), figure 2.7.1: RLBench: success rate (18 tasks, 100 demo/task). Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion