Skip to content

Releases: StanfordVL/BEHAVIOR-1K

BEHAVIOR-1K v3.7.1

12 Sep 23:38
@wensi-ai wensi-ai
v3.7.1
79f0f5f
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
BEHAVIOR-1K v3.7.1 Latest
Latest

In this release, we fixed various bugs and added more features/supports for the 1st BEHAVIOR challenge. Visit https://behavior.stanford.edu/ for more information!

What's Changed

  • Removed grasp_left and grasp_right from robot proprioception and updated the demo dataset accordingly.
    • Now, the dimension of observation.state in the dataset is 256 instead of 258
  • Updated task-instances dataset, included 2 more metadata files for data replay
  • Fixed library errors (av, gspread, pandas, cffi) during installation
  • Added RGBLowResWrapper and set it to be the default wrapper during evaluation rollout
  • Added GOP-level randomization to BehaviorLeRobotDataset for faster data access during training
  • Added point cloud modality to vision sensor
  • Updated camera relative pose to base implementation, changed n_render_iterations to 1 during evaluation rollout
  • Fixed bugs in demo replay script
  • Added NO_OMNI_LOG macro
  • Updated challenge rules and documentations:
    • Changed evaluation rollout number from 20 * 5 to 10 * 1
    • Specified robot base global pose is not allowed in standard track
    • Added video recordings as part of the submission

To migrate from v3.7.0 to v3.7.1

  • If you already downloaded the demo dataset, please update this by syncing local dataset with HF 
    ds = BehaviorLeRobotDataset(
    repo_id="behavior-1k/2025-challenge-demos",
    root=PATH_TO_DATASET_ROOT,
    tasks=LIST_OF_TASK_NAMES,
    download_videos=False,
    local_only=False,
    force_cache_sync=True
)
    Specify task names under tasks, or set to None if you want to update all tasks.
    If you are only downloading the dataset, you can safely exit the program after seeing something like 
    Generating train split: 23346708 examples [03:25, 112135.76 examples/s]
  • Pull the latest code from v3.7.1 (79f0f5), then within the behavior conda environment run: 
    ./setup.sh --bddl --omnigibson --dataset

Full Changelog: v3.7.0...v3.7.1

Assets 2
Loading

BEHAVIOR-1K v3.7.0

02 Sep 20:08
@cgokmen cgokmen
v3.7.0
44186f1
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
BEHAVIOR-1K v3.7.0

In this release, we add support for Isaac Sim 4.5, and support all features for the BEHAVIOR challenge including teleoperation and evaluation. Visit https://behavior.stanford.edu for more information!

What's Changed

Full Changelog: v1.1.1...v3.7.0

Contributors

cgokmen, ChengshuLi, and 3 other contributors
Loading

OmniGibson v1.1.1

04 Oct 05:21
@cgokmen cgokmen
v1.1.1
d15de67
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
OmniGibson v1.1.1

This release contains the below bugfixes to bugs affecting installation, particle systems, action primitives, profiling, and segmentation images:

What's Changed

Full Changelog: v1.1.0...v1.1.1

Contributors

cgokmen, ChengshuLi, and 2 other contributors
Loading

OmniGibson v1.1.0

02 Oct 03:29
@cgokmen cgokmen
v1.1.0
032c9b2
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
OmniGibson v1.1.0

We are excited to announce the latest release of OmniGibson, bringing significant enhancements to our simulation platform for embodied AI research and development.

Highlights

  • Fast Installation with pip: You can now install OmniGibson and Isaac Sim via pip! For detailed
    installation instructions, please visit our official documentation.
  • Multiple/Parallel Environments: Support for reinforcement learning with optimized performance
    by running multiple environments at once as in Isaac Gym.
  • Data Collection and Playback: You can now use our data collection environment wrappers for demonstration data.
  • Primitive Skill Library: We now provide more robust action primitives and faster IK, with primitives
    now covered under integration tests for reliability.
  • Torch Integration: OmniGibson now works exclusively with torch tensors instead of numpy arrays, in
    anticipation of full GPU-based simulation and data pipeline.
  • cuRobo Integration: OmniGibson now supports some accelerated motion planning capabilities from curobo.
    This integration will be extended in the near future to fully remove the OMPL dependency in primitives.
  • Scene Graph Generation: OmniGibson can now generate scene graphs, compatible with single or multiple agents.
  • New Robots: We support a number of new robots and provide improvements for our existing robots.
  • Documentation Overhaul: Detailed OG modules and tutorials are now provided in the documentation.

Breaking Changes

  • OmniGibson now uses torch instead of numpy as is numerical computation backend, in anticipation of
    full GPU-based simulation features that we plan to include in the next release. This means that users
    now need to provide torch tensors when calling APIs instead of numpy arrays.
  • OmniGibson is no longer compatible with Isaac Sim 2023.1.1 - we recommend upgrading to 4.1. OmniGibson
    is not yet compatible with Isaac Sim 4.2.
  • Local pose APIs have been removed and unified into the set_position_orientation and get_position_orientation
    APIs with the frame argument.
  • Absolute prim paths may no longer be provided when creating objects and prims. You will need to provide
    a relative prim path which will be applied on top of the object's scene's prim, in order to support
    multiple environments when necessary.
  • Inverse Kinematics controllers have been updated to use a local approximation ("differential IK")
    instead of solving the full IK problem on every step, providing significantly better speed especially
    in multi-environment contexts.

We encourage all users to upgrade to this latest version to take advantage of these improvements and new features. As always, we welcome your feedback and contributions to help make OmniGibson even better.

Merged PRs

Read more

Contributors

cgokmen, ChengshuLi, and 5 other contributors
Loading

OmniGibson v1.0.0

18 Mar 06:57
@cgokmen cgokmen
v1.0.0
282adda

Choose a tag to compare

View all tags
OmniGibson v1.0.0

We are happy to announce the first full release of BEHAVIOR-1K with 1,004 pre-sampled task instances now fully functional! See docs at https://behavior.stanford.edu/omnigibson/getting_started/installation.html

Highlights

  • Expanded interactive scenes
  • Enhanced object interaction
  • Teleoperation System
  • Advanced feature support: action primitives, transition rules, etc.

Change log

  • Assets:

    • We now offer all 50 interactive traversable scenes
    • Fillable object meshes have been updated with tighter, more accurate fillable volumes
    • Comprehensive meta link annotations
    • Full sliceable synset coverage
    • 1004 pre-sampled task initial conditions from the B1K activity suite

  • Simulation:

    • Operational Space Controller support
    • New teleoperation system supporting OpenXR (Meta Quest), SteamVR (HTC VIVE), SpaceMouse, and Keyboard teleoperation
    • Assisted grasping support
    • More robust cloth generation
    • Support for external sensors
    • Controller logic has been significantly refactored to properly handle staggered action and physics timestep differences
    • Controller performance has been optimized, and gravity compensation has been fixed
    • Lazy import: importing OmniGibson no longer immediately launches the simulator
    • Poses in USD, Fabric, and PhysX are now synchronized without taking simulation steps
    • Full transition rule support
    • Deterministic global labels for semantic segmentation and consistent labels for instance segmentation
    • Segmentation support for particle systems
    • More accurate bounding box computation
    • Improved support for non-uniformly scaled meshes
    • Improved task sampling support
    • Action primitives support

  • Usability:

    • Comprehensive documentation, including modules, tutorials, and updated examples
    • Logging has been pruned for a cleaner output.
    • Profiling support
    • A Colab notebook for running OmniGibson without local installation is now available

Full Changelog: v0.2.1...v1.0.0

Loading

OmniGibson v0.2.1

29 Aug 22:49
@cremebrule cremebrule
v0.2.1
aa8b870
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
OmniGibson v0.2.1

Bug Fixes

  • Fix task sampling errors (#253)
  • Improved dataset quality (better annotated metalinks)
  • Improved transition rule useability
  • Fix Tiago floating issue
  • Improved sampling robustness
Loading

OmniGibson v0.2.0

07 Aug 06:42
@cremebrule cremebrule
v0.2.0
452b28f
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
OmniGibson v0.2.0

BEHAVIOR-1K v0.2.0 beta release of OmniGibson, with a significantly expanded dataset and broad code improvements. See docs at https://stanfordvl.github.io/OmniGibson/getting_started/installation.html

Known Issues

  • The latest version of Isaac Sim (2022.2.1) may cause segfault when loading scenes with certain OmniGibson configurations

Highlights

  • Revamped dataset
  • Expanded object states suite and improved cloth features

Change log

  • Assets:

    • 22 Scenes, including all 15 iGibson scenes and 7 new large-scale scenes (house_single_floor, house_double_floor_lower, house_double_floor_upper, grocery_store_cafe, restaurant_brunch, office_large, Pomaria_0_garden)
    • Improve collision mesh quality
    • Many new object assets, with added support for full OmniGibson object state features
    • 300+ systems, including fluid, granular, and visual-type particles
    • 600 pre-sampled task initial conditions from the B1K activity suite

  • Simulation:

    • Added object states:
      • Contains
      • Draped
      • Covered for attaching visual particles to cloths
    • Refactored systems
      • New particle type: MacroPhysicalParticleSystem, for physical, "large" particles that are in the ~10-50 instance range (e.g.: small fruits, nuts, etc.)
    • Refactored BehaviorTask
      • Refactored online task sampling scheme for better robustness and generalizability with BDDLSampler
    • Dynamic clothification at runtime to allow users to input arbitrary objects to be converted to cloths
    • Added RecipeRule transition rules to enable programmatic, modular rule specifications
    • Highlighted Bug Fixes:
      • Fix scaling of prismatic joints
      • Fix bugs related to assisted-grasping
      • Tune cloth hyperparameters for better physical behavior

  • Usability:

    • Improved standardization and docstrings to improve code interpretability / extensibility
    • Improved conda environment installation scripts
    • Updated installation documentation
Loading

OmniGibson v0.1.0

10 Apr 21:51
@wensi-ai wensi-ai
v0.1.0
c38cfdd
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
OmniGibson v0.1.0

BEHAVIOR-1K beta release of OmniGibson, with new features and significant improvements over installation, stability, and performance. See docs at https://behavior.stanford.edu/omnigibson/

Known Issues

  • The latest version of Isaac Sim (2022.2.1) may cause segfault when loading scenes with certain OmniGibson configurations

Highlights

  • Cross-platform compatibility and better installation robustness
  • All demo scripts work error-free
  • Improved rendering quality and simulation speed

Change log

  • Assets:
    • Included 5 iGibson scenes
    • Improved collision mesh quality
    • Fixed lights in all existing scenes
  • Simulation:
    • Revamped object states:
      • Attached to AttachedTo
      • FluidSource / FluidSink to ParticleSource / ParticleSink
    • Added unit tests for object states
    • Added demo scripts for Dicing and ParticleSource / ParticleSink
    • Updated RL standalone training example
  • Usability:
    • Added support for Windows
    • Improved conda environment installation scripts
    • Updated installation documentation
    • Cleaned up logging and spurious omni warnings / errors
Loading

OmniGibson v0.0.6

08 Apr 04:03
@wensi-ai wensi-ai
v0.0.6
c77de3e
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
OmniGibson v0.0.6 Pre-release
Pre-release

BEHAVIOR-1K beta release of OmniGibson.

This release has been deprecated due to bugs. Please use v0.1.0 instead.

Loading

OmniGibson v0.0.5

21 Feb 01:00
@cgokmen cgokmen
v0.0.5
915add3
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
OmniGibson v0.0.5

BEHAVIOR-1K beta release of OmniGibson, with bugfixes and compatibility with Isaac Sim 2022.2.0 as well as updates to Docker initialization. See docs at https://behavior.stanford.edu/omnigibson/

Loading