Multi Tasking

Intro

At the moment this is done rather basically. A number of instances of the different environments are created and they generate data in parallel for the learner to train on. One constraint of doing things this way is that the number of threads used during training should the same as the number of tasks. If there are two tasks, use 2 threads... if 3 tasks use 3, etc..

Example:

Give a list of the environments to run. The list

{
    "comment__": "Config file for the simulator",
"sim_config_file": ["./args/genBiped2D/biped2dfull_flat_with_terrain_features.txt", "./args/genBiped2D/biped2dfull_incline_with_terrain_features.txt", "./args/genBiped2D/biped2d_full_steps.txt", "./args/genBiped2D/biped2d_full_slopes_with_terrain_features.txt", "./args/genBiped2D/biped2d_full_gaps_with_terrain_features.txt"],
    "comment__": "A different config file for the forward dynamics simulator",
"forwardDynamics_config_file": ["./args/genBiped2D/biped2dfull_flat_with_terrain_features.txt", "./args/genBiped2D/biped2dfull_incline_with_terrain_features.txt", "./args/genBiped2D/biped2d_full_steps.txt", "./args/genBiped2D/biped2d_full_slopes_with_terrain_features.txt", "./args/genBiped2D/biped2d_full_gaps_with_terrain_features.txt"],
"num_available_threads": 4,
	"comment__": "This overrides all of the envs to be a single one from the list of sim environments",
"override_sim_env_id": 4
}