Create custom workspace

MyGym toolkit is designed to be modular and so is the environment GymEnv. It includes several Workspaces among which you can easily alternate to add variability to your training. This tutorial shows how to create and add your custom workspace into Gym and train your task there.

The definition of GymEnv is here. There you can find the definitions of existing workspaces in the self.workspace_dict dictionary.

Choose a key for your workspace and add new value with its definition into this self.workspace_dict:

self.workspace_dict =  {'myworkspace':  {'urdf': 'myworkspace.urdf', 'texture': 'mytexture.jpg',
                                        'transform': {'position':[1.5, -1.0, -1.05], 'orientation':[0.0, 0.0, -0.5*np.pi]},
                                        'robot': {'position': [0.0, 0.0, 0.0], 'orientation': [0.0, 0.0, 0.5*np.pi]},
                                        'camera': {'position': [[0.56, -1.71, 0.6], [-1.3, 3.99, 0.6], [-3.43, 0.67, 1.0], [2.76, 2.68, 1.0], [-0.54, 1.19, 3.4]],
                                                    'target': [[0.53, -1.62, 0.59], [-1.24, 3.8, 0.55], [-2.95, 0.83, 0.8], [2.28, 2.53, 0.8], [-0.53, 1.2, 3.2]]},
                                        'boarders':[-0.7, 0.7, 0.3, 1.3, -0.9, -0.9]},
                                        ...}

You need to provide name of the urdf definition file of the model of your workspace and optionally a path to the texture you want to apply on the model. The urdf should contain correct position, orientation and scale of the model, to be loaded into the Gym properly. You will need two urdf definition files - one with visual and collision definitions and one with visual definition only. Store these files to the proper locations: collision and visual respectively. This way you ensure that the collision detection will work during training in your custom workspace and that a visual copy of your workspace will be visible while training in another.

To make things easier, we recommend to place the origin of the Cartesian coordinate frame to the position and orientation known with respect to the workspace. Use the transform value for this purpose.

Define position and orientation within your workspace, where the robot should be initialized. If you applied transform, use the new coordinates now.

Define position and target position of cameras accordingly. The cameras are used for visualization and image rendering for vision modules.

Lastly, define volume within your workspace, where task objects are allowed to occur. You can also adjust this parameter later using object_sampling_area parameter.

To start your custom training, set your new workspace keyword as the workspace parameter in the config file:

"workspace":"myworkspace",

or in the command line:

python train.py --workspace=myworkspace