Stanford
ACT↗
ACT is a Baseline from Stanford, released April 23, 2023. It ranks #34 of 36 overall on the Robot Intelligence Index, with a rating of 1108. Its best result is #6 of 9 on VLA-Replica. Action Chunking with Transformers: CVAE transformer policy for bimanual ALOHA; standard single-task baseline.
Release date: Apr 23, 2023
- Developer
- Stanford
- Type
- Baseline
- Model weights
- Open
- License
- MIT
- Parameters
- 0.08B
- Pricing
- Self-hosted
- Benchmarks
- 4
- Paper
- Read ↗
Overall rank
#34of 36
Rating 1108 · range 27–35
Real robots rank
#54of 56
Rating 1194 · range 24–48
Simulation rank
#45of 49
Rating 1064 · range 36–46
Rating
1108± 13
10601989
Cost / task · est.
$0.030
$0.0087$17.74
Time / task · est.
33s
9.5 s1 h 22 min
Rank on each benchmark
Per task attempt; the tiles show the median across benchmarks. Self-hosted compute is priced at one H100 at $3.29 per hour (Lambda on-demand, Oct 2026).
Other models from Stanford
Inference
Small CVAE transformer (~80M params); predicts 100-step chunks for 50 Hz ALOHA control. OFT paper measured 58 ms latency on A100 (ALOHA, 3 images).