Simulation Environments

A simulation environment is a composed MuJoCo model: a human musculoskeletal (MSK) model, an assistive device, and a terrain and the same definition drives both the reflex Controller Optimization and the Reinforcement Learning pipelines. See Defining an Environment.

Composed environments across varied terrain (left) and a target-velocity field over a tiled course (right).

How an environment is composed

Three sibling packages supply the parts, and MyoAssist composes them into one useable model:

• myo_sim provides the human MSK models and their meshes.
• assist_sim holds the device configurations and the combination pipeline. It composes one MSK and one device into a single runnable model.
• myoassist.terrains provides the scene (ground, heightfields, and associated velocities). The assist_sim output is model-only, with no ground.
• myoassist consumes the composed model for Controller Optimization and Reinforcement Learning.

The result is a single MuJoCo model that the Controller Optimization and Reinforcement Learning pipelines step.

Environment composition pipeline.

In this section

• MSK Models: the human models: myolegs26, myolegs22, myolegs, and myofullbody.
• Assistive Devices: the device catalog, the gait-assistive and upper-body environments, device configuration, and how to add, prepare, and export a device.
• Terrains: tile types, terrain configuration, and velocity maps.

---

Table of contents

• Musculoskeletal Models
• Assistive Devices
• Terrains