Roboschool

Roboschool (https://github.com/openai/roboschool) provides several environments for controlling robots in simulation. It was released by OpenAI and the environments have the same interface as the OpenAI Gym environments that we have been using in this book. The Gym's MuJoCo-based environments offer a rich variety of robotic tasks, but MuJoCo requires a license for use after the free trial. Roboschool provides eight environments that quite closely match the MuJoCo ones, which is a good news as it offers a free alternative. Apart from these eight environments, Roboschool also offers several new and challenging environments.

The following table shows a quick comparison between the MuJoCo Gym environments and the Roboschool environments:

Brief description MuJoCo environment Roboschool environment

Make a one-legged 2D robot hop forward as fast as possible

Hopper-v2

RoboschoolHopper-v1

Make a 2D robot walk

Walker2d-v2

RoboschoolWalker2d-v1

Make a four-legged 3D robot walk

Ant-v2

RoboschoolAnt-v1

Make a bipedal 3D robot walk forward as fast as possible without falling

Humanoid-v2

RoboschoolHumanoid-v1

 

A full list of environments that are available as part of the Roboschool library, with their state and action spaces, is provided in the following table for your quick reference:

Env ID
Roboschool env
obs space
action space
RoboschoolInvertedPendulum-v1 Box(5,) Box(1,)
RoboschoolInvertedPendulumSwingup-v1 Box(5,) Box(1,)
RoboschoolInvertedDoublePendulum-v1 Box(9,) Box(1,)
RoboschoolReacher-v1 Box(9,) Box(2,)
RoboschoolHopper-v1 Box(15,) Box(3,)
RoboschoolWalker2d-v1 Box(22,) Box(6,)
RoboschoolHalfCheetah-v1 Box(26,) Box(6,)
RoboschoolAnt-v1 Box(28,) Box(8,)
RoboschoolHumanoid-v1 Box(44,) Box(17,)
RoboschoolHumanoidFlagrun-v1 Box(44,) Box(17,)
RoboschoolHumanoidFlagrunHarder-v1 Box(44,) Box(17,)
RoboschoolPong-v1 Box(13,) Box(2,)
..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset
18.218.129.100