Choose a version to compare

  • LeRobot ACT · Tridex 3Cam ResNet 100K 18_11_2
  • ME297SJSU2026 ACT SO-101 Dial Pick-and-Place Mix v3 DAgger2
  • ME297SJSU2026 SO-101 v3 DAgger3
  • kiteml/trossen-bin-picking-sim-act-cup-all-90k
  • SiMa.ai ACT SO-101 PickPlace
  • LeRobot ACT
  • Kite Trossen Cup Real ACT
  • Kite Trossen Sim ACT Cup Baseline · 30k
  • Kite Trossen Sim Cup All ACT 30k
  • Kite Trossen Cup Sim Baseline ACT

Action Chunking with Transformers policy for fine-grained robot manipulation. Original paper first submitted: 2023-04-23. Co-developed by Stanford University, UC Berkeley and Meta.

Overview

Model type
Action policy
Architecture
Transformer
Listed versions
LeRobot ACT · Tridex 3Cam ResNet 100K 18_11_2 / ME297SJSU2026 ACT SO-101 Dial Pick-and-Place Mix v3 DAgger2 / ME297SJSU2026 SO-101 v3 DAgger3 / kiteml/trossen-bin-picking-sim-act-cup-all-90k / SiMa.ai ACT SO-101 PickPlace / LeRobot ACT / Kite Trossen Cup Real ACT / Kite Trossen Sim ACT Cup Baseline · 30k / Kite Trossen Sim Cup All ACT 30k / Kite Trossen Cup Sim Baseline ACT
Related robots
2

Model history

November 2026

SiMa.ai upgrades the ACT SO-101 deployment bundle to Palette SDK 3.0

Real-world useOfficial announcement

SiMa.ai compiled this task-specific ACT policy for the Modalix DevKit. An SO-101 follower uses two USB camera views to pick up a gray eraser and place it in a white basket. The public BF16 bundle comes from step 100,000, contains six compiled graphs and predicts 100 six-dimensional actions per chunk.

Previously reported

The previous compiled bundle was posted on October 23 UTC (October 24 in Taipei). It used the reference toolchain ModelSDK 2.1.3, Neat 0.5.0 and sima_lmm 0.5.0.

This announcement

2026-11-19 08:18:57 (UTC+08:00) Palette SDK 3.0.0 · Model Compiler 3.0.0 · LLiMa 0.6.0 act-so101-grey-eraser-pickplace-100k-bf16-219727d9 SiMa.ai reports a median model latency of 13.5 ms, cosine similarity of 0.999992 and maximum absolute error of 0.0147 over 20 fixture plans after three warm-ups. The reference inputs were replayed without moving the robot.

The release contains compiled ELFs, reference fixtures and runtime assets. The training checkpoint and dataset are outside this distribution. Changes in geometry, lighting or calibration can affect behavior. Physical task success rate is unreported.

April 2023

Original ACT method paper becomes available

New resourcesOfficial announcement

The original ACT paper introduces action chunking for fine-grained bimanual manipulation.

This announcement

A public research report describes a conditional generative policy over action sequences using visual observations and joint positions.

This is the original method paper’s first arXiv publication, not a release date for the later LeRobot implementation. No checkpoint or port-specific evaluation is inferred.

View all important updates →

Papers & reports (1)

Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware

Paper available

arXiv: 2304.13705
Authors & affiliations (4)

Author order: Original order

  1. Tony Z. Zhao

    Not provided

  2. Vikash Kumar

    Not provided

  3. Sergey Levine

    Not provided

  4. Chelsea Finn

    Not provided

First published
2023-04-23
Latest paper revision
v1 · 2023-04-23
Venue
RSS 2023

Covered model versions

  • act · LeRobot ACT

    Foundational ACT method implemented by LeRobot; not an exact LeRobot implementation release report or a benchmark for an unspecified trained checkpoint.

Paper ↗PDF ↗

Official resources

Submit a source or correction