IKEA Furniture Assembly Environment

We are announcing the launch of the IKEA Furniture Assembly environment as a first-of-its-kind benchmark for testing and accelerating the automation of physical assembly processes. The IKEA Furniture Assembly environment will be a stepping stone towards a new era of robots that can understand, plan, and skillfully execute precise physical manipulations, a far departure from the traditional robots of today.
The current release is a beta release. We welcome any feedback using this form.
Beta v1.0 (December, 2019)
10 furniture models, 6 background scenes 80+ furniture models, programmatic background scenes
Baxter, Sawyer, Cursor agents More robots (e.g. Jaco, Fetch)
IK and Impedance Control Support for 3D controllers
RL/IL benchmarks

Furniture Assembly

Complex long-horizon manipulation task

Diversity

80+ furniture models, customizable background, lighting and textures

Robots

Baxter, Sawyer, and more

Easy to Use

OpenAI Gym interface for imitation / reinforcement learning

Realistic Robot Simulation

Sawyer
Baxter
Cursor
  • Support for Baxter and Sawyer with IK, impedance, and torque controls
  • More robots and grippers such as Jaco, UR, and Fetch coming soon!
  • Provide Cursor agent as an abstraction

Various Environments

Diverse lighting, textures, and background
  • Variability in furniture, textures, physics, lighting, background, and more
  • Uses Unity3d for realistic visualization

80+ IKEA Furniture Models

Diverse IKEA furniture
  • Over 80 furniture models: chairs, tables, cabinets, bookcases, desks, shelfs, and TVs
  • Created following official IKEA manuals with minor simplifications in carving and screws
Examples of furniture models

Available Observations

Scene
Depth
Segmentation
Goal
  • Reinforcement learning and imitation learning for complex manipulation tasks
  • Sim-to-real for real robot object manipulation
  • Computer vision tasks (object pose estimation, scene understanding, instance segmentation, etc)

Limitations and Future Work

While we worked hard to provide a comprehensive furniture assembly environment, there are many potential avenues for extension.

  • Additional robots: Fetch, UR and grippers
  • Realistic part attachment: peg insertion, screwing, nailing mechanisms
  • Tool use: using a screwdriver or a hammer
  • Multi-agent assembly: using multiple robots to assemble furniture
  • Supervision: leverage verbal and visual instructions or demonstrations
  • Identical parts: Parts can be interchangable

Citations

                      @article{lee2019ikea,
  title={{IKEA} Furniture Assembly Environment for Long-Horizon Complex Manipulation Tasks},
  author={Lee, Youngwoon and Hu, Edward S and Yang, Zhengyu and Yin, Alex and Lim, Joseph J},
  journal={arXiv preprint arXiv:1911.07246},
  year={2019},
  url={https://clvrai.com/furniture},
}
                    

Questions and Comments

For bug reports, feedback, or comments, please fill out this form, or create a github issue, or email Youngwoon Lee (lee504@usc.edu).