Environments

LyceumMuJoCo comes with a variety of environments:

  • Lyceum Suite: A collection of Lyceum-custom environments.
  • Gym: Ports of environments from OpenAI's gym.
  • dm_control: Ports of environments from DeepMind's dm_control

We highly encourage that users get familiar with the source codes of these environments and modify them as you see fit. We also hope that they serve as inspiration for creating new, interesting environments. As always, if you make something cool we'd gladly welcome a pull request to incorporate it into Lyceum!

Lyceum Suite

PointMass

LyceumMuJoCo.PointMassType
struct PointMass{S<:MJSim, O} <: LyceumMuJoCo.AbstractMuJoCoEnvironment

PointMass is a simple environment useful for trying out and debugging new algorithms. The task is simply to move a 2D point mass to a target position by applying x and y forces to the mass.

Spaces

  • State: (13, )
  • Action: (2, )
  • Observation: (6, )

ArmHandPickup

LyceumMuJoCo.ArmHandPickupType
struct ArmHandPickup{S<:MJSim, O<:Shapes.MultiShape} <: LyceumMuJoCo.AbstractMuJoCoEnvironment

Pickup a block using a robot arm modeled after the Modular Prosthetic Limb developed by the Applied Physics Laboratory, The Johns Hopkins University.

Spaces

  • State: (106, )
  • Action: (36, )
  • Observation: (19, )

Gym

SwimmerV2

LyceumMuJoCo.SwimmerV2Type
mutable struct SwimmerV2{SIM<:MJSim, S<:Shapes.AbstractShape, O<:Shapes.AbstractShape} <: LyceumMuJoCo.AbstractMuJoCoEnvironment

This task involves a 3-link swimming robot in a viscous fluid, where the goal is to make it swim forward as fast as possible, by actuating the two joints. The origins of task can be traced back to Remi Coulom's thesis: "Reinforcement Learning Using Neural Networks, with Applications to Motor Control"

  • State: (17, )
  • Action: (2, )
  • Observation: (8, )

Walker2DV2

HopperV2

dm_control

CartpoleSwingup