Open Source RoboticsRobots, software & learning

robomimic

Software

Robot-learning-from-demonstration framework with HDF5 dataset tooling, offline learning algorithms and version-specific simulation datasets. [1]

Installation & first result →Official websiteGetting started

For
Researchers comparing imitation and offline reinforcement-learning methods [1]
Source rights
Some licenses are unresolved. Review licenses
Unknowns
compute requirements not fully established. Review gaps

Source researched · Not tested by us
Sources checked · Research details

Licenses & related hardware
License
MIT · license not fully assessed for documentation · Details by part

Requirements · Research details and dates

On this page

Overview

robomimic provides algorithms and data tooling for learning from robot demonstrations. The reviewed v0.5.0 framework supports configuration-driven training and a common HDF5 trajectory format. Its value for a comparison depends on preserving the dataset, simulator and preprocessing choices together. An installed training package alone does not reproduce an earlier experiment. [1] [2] [7]

Is this for you?

Workflows
  • Download or collect demonstrations
  • Convert and inspect HDF5 data
  • Choose train-validation trajectory filters
  • Train policies with a dataset and configuration
  • Replay simulated states to generate selected observations

[7] [8]

Known limitations
  • Newer robosuite datasets may not reproduce original CoRL2021 learning results exactly
  • Raw converted robosuite datasets still need observation extraction
  • Images require uint8 channel-last format; actions should be normalized to [-1,1]
  • Real-robot datasets are not included in the linked Hugging Face collection

[7] [8] [9]

What you need

Operating system
  • Linux
  • macOS

[5]

Languages / runtimes
  • Python

[5] [6]

Dependencies
  • PyTorch and torchvision
  • HDF5 tooling via h5py
  • huggingface_hub ==0.23.4
  • transformers ==4.41.2
  • diffusers ==0.11.1
  • Dataset-specific simulation dependencies

[5] [6]

Installation & first result

Open the upstream getting-started guide [5] · Project documentation [4] [5]

Installation

Select an isolated Python/PyTorch environment, install the pinned source, and follow the dataset-specific simulator instructions before the debug training example.

Guide recommends Python 3.8.0 and shows PyTorch 2.0.0; no resolved environment tested here.

[5] · Editorial assessment

Hardware, models & data

Interfaces
  • HDF5 demonstration format
  • Configuration-driven training script
  • Dataset conversion, inspection and train-validation splitting utilities

[7] [8]

Inputs and outputs

HDF5 trajectories contain actions, rewards, termination signals, observations and environment metadata; training consumes a dataset path and experiment configuration.

[7]

Project notes

Installation and experiment reproduction are separate checks

The pinned installation guide names Linux and macOS, recommends Python 3.8.0 and shows PyTorch 2.0.0 with torchvision 0.15.1. Package metadata has a much broader Python declaration and also pins several learning dependencies, including transformers 4.41.2 and diffusers 0.11.1. These are source-level instructions, not a resolved environment tested here. Use an isolated environment and record the versions that actually install before treating it as reproducible. [5] [6]

The guide's small installation check is examples/train_bc_rnn.py --debug. Dataset use can require additional simulators. For the released robosuite demonstration route, the documentation points to robosuite v1.5.1. That is a more specific choice than its general statement of robosuite v1.2-or-newer compatibility. [5]

The dataset guide explains why the distinction matters: the original CoRL 2021 data used the older offline_study branch and mujoco-py. Newer datasets use robosuite v1.5.1 with the official MuJoCo bindings, and upstream explicitly says learning results may not match the original results exactly. Older reproduction routes are documented through robomimic v0.2.0 or v0.3 for their respective dataset generations. Do not silently substitute one generation when reporting a historical comparison. [9]

Inspect the data before training

A compatible HDF5 file contains trajectory groups and environment metadata, with actions, observations, rewards and termination signals. Images are stored as unsigned 8-bit arrays in channel-last order. Actions should be normalized between minus one and one; the documented dataset inspection script checks their range. An HDF5 extension alone therefore does not establish compatibility. [7]

The robosuite conversion route changes a raw demonstration file in place. After that conversion, the data still lacks observations, rewards and termination signals needed for training. The separate state-to-observation step reconstructs selected observations and rewards from the recorded MuJoCo states. Preserve a copy of raw data before an in-place conversion; this is an editorial precaution for keeping the original evidence. [8]

Generate train-validation splits before extracting observations. Upstream recommends that ordering so every derived HDF5 file inherits the same trajectory filters. Otherwise, comparisons between image and low-dimensional observations can inadvertently use different splits. The framework exposes filter keys for selecting training and validation trajectories explicitly. [7] [8]

Observation extraction also chooses camera names, resolution, depth, reward density and termination semantics. Its examples distinguish marking success from marking only the trajectory end. Compression trades smaller storage for decoding work; omitting next observations is appropriate only for algorithms that do not require them. Preserve these choices with the experiment configuration rather than treating every derived dataset as equivalent. [8]

What we checked

Source research only; no installation, physical build or performance test performed. No affiliate links added.

Not yet established

  • Requirements: host computer
  • A step-by-step setup summary
  • Version-specific hardware compatibility (ecosystem tags alone do not establish it)

Updates

Upstream last updated
Jun 27, 2025 [2]
Latest release
Jun 27, 2025 [2]
Profile updated
Oct 1, 2026
Sources last checked
Oct 1, 2026
Next review
Oct 31, 2026
Reviewed revision
ae5799f0fae05c4559ee1f9645b0f77eb5251929 (release v0.5.0) · software repository

Source files and licenses

Available project sources support research coverage; complete build and commercial qualification remain separate.

LayerAvailability / license
softwareavailable · MIT
Framework license; dependencies and externally hosted datasets are separate artifacts. [3]
datasetsavailable · MIT
Maintainer dataset card reports MIT for this hosted collection. Dataset payloads not downloaded or validated. This does not cover every supported external dataset or real-robot release. [9] [10]
documentationavailable · License not fully assessed
Documentation source present; separate documentation-wide terms not established. [4] [5]

Framework and demonstration-data profile; no specific pretrained checkpoint selected.

Keep source and dataset rights separate

The framework's root license is MIT. The separately inspected maintainer-hosted Hugging Face dataset card also labels its collection MIT, but that collection is only part of the broader supported-data ecosystem. The dataset guide says its real-robot releases are not hosted there. Neither license statement establishes the terms of every external dataset supported by the framework. [3] [9] [10]

This review inspected source and documentation and preserved the dataset card revision. It did not download the HDF5 payloads, resolve an installation, replay states, train a policy or measure a reproduction result. [2] [10]

References

Each source links to the exact revision we read. “Used for” shows where this page relies on it.

Correction guidance · Discussion is not enabled · Profile format 0.6.0-draft.1

Change history