Welcome to MIMo’s documentation!

MIMo is a library for the research of cognitive development and multi-sensory learning. We follow the API of Gymnasium environments, using MuJoCo for the physics simulation.

Note

This project is under active development.

Contents

• User Guides
  • Installation
  • Creating your own experiment
  • Sensory modules
• API
  • MIMoEnv base class
  • Actuation Models
  • Sample environments
  • Proprioceptive system
  • Touch system
  • Vestibular system
  • Vision system
  • Utility functions
• Changelog
  • Version 1.1.0 (The Big Migration)
  • Version 1.0.0 (Initial release)
  • Version 0.1.0 (Paper release)

MIMo is a simple, childlike robot physically simulated using MuJoCo. He has multiple sensory modalities including touch, vision, proprioception and a vestibular system.

There are two physical models, which are identical except for the hands and feet. The base version uses mitten-like hands with only a single finger. The feet also only have one toe. The v2 version has fully modelled five-fingered hands based on the Adroit hand and feet with two toes. There is a demo environment for the v2 version available: MIMoV2DemoEnv.

We also offer two different actuation models. See the Actuation API for more details.

The sensory modules can be configured using parameter dictionaries. Defaults exist for all of them, but note that the parameters for the sensor modules are specific to the model used. The touch parameters for example expect that relevant MuJoCo bodies exist in the scene and will cause an error otherwise. Since the base and the v2 version of MIMo have different bodies they also require different touch parameters.

Gym environments provide the interfaces to interact with the environments. See the gym documentation for examples. Sample environments and simple demo scripts can be found in the Samples section.