graph LR
Simulation_Core["Simulation Core"]
Environment_Task_Definition["Environment & Task Definition"]
Asset_Scene_Management["Asset & Scene Management"]
Robot_Agent_System["Robot & Agent System"]
Data_Trajectory_Management["Data & Trajectory Management"]
Learning_Algorithms_Baselines["Learning Algorithms & Baselines"]
Environment_Task_Definition -- "configures and initializes" --> Simulation_Core
Asset_Scene_Management -- "provides elements to" --> Simulation_Core
Simulation_Core -- "provides observations to" --> Robot_Agent_System
Robot_Agent_System -- "sends actions to" --> Simulation_Core
Learning_Algorithms_Baselines -- "collects experience from" --> Simulation_Core
Simulation_Core -- "provides feedback to" --> Learning_Algorithms_Baselines
Learning_Algorithms_Baselines -- "stores data in" --> Data_Trajectory_Management
Data_Trajectory_Management -- "provides data to" --> Learning_Algorithms_Baselines
Data_Trajectory_Management -- "can replay episodes in" --> Simulation_Core
click Simulation_Core href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/ManiSkill/Simulation_Core.md" "Details"
click Environment_Task_Definition href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/ManiSkill/Environment_Task_Definition.md" "Details"
click Asset_Scene_Management href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/ManiSkill/Asset_Scene_Management.md" "Details"
click Robot_Agent_System href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/ManiSkill/Robot_Agent_System.md" "Details"
click Data_Trajectory_Management href "https://github.com/CodeBoarding/GeneratedOnBoardings/blob/main/ManiSkill/Data_Trajectory_Management.md" "Details"
The ManiSkill project is architected around a robust simulation core that interacts with various specialized components to enable complex robotic manipulation tasks. The Simulation Core forms the heart of the system, managing physics, scenes, and rendering. Asset & Scene Management feeds into this core by providing the necessary 3D models and environmental structures. Robotic agents, defined within the Robot & Agent System, interact with the Simulation Core by sending actions and receiving observations. The Environment & Task Definition component configures the simulation, setting up specific tasks and their parameters, which are then executed within the Simulation Core. Finally, Learning Algorithms & Baselines leverage the simulation for training, collecting and storing data via Data & Trajectory Management, and using this data to refine agent policies. This modular design facilitates clear separation of concerns, allowing for independent development and optimization of each component, which is crucial for both documentation and visual diagram generation.
Simulation Core [Expand]
The central physics simulation engine, scene manager, and rendering interface.
Related Classes/Methods:
mani_skill.envs.sapien_envmani_skill.envs.scenemani_skill.sensors.cameramani_skill.envs.wrappers.gym_wrapper
Environment & Task Definition [Expand]
Defines specific simulation environments and tasks, including goals, reward functions, and initial scene configurations.
Related Classes/Methods:
Asset & Scene Management [Expand]
Manages the loading, creation, and manipulation of 3D assets to construct simulation scenes.
Related Classes/Methods:
mani_skill.utils.building.actor_buildermani_skill.utils.building.articulation_buildermani_skill.utils.building.urdf_loadermani_skill.utils.assets.data
Robot & Agent System [Expand]
Defines robotic agents, their physical models, and control mechanisms for interacting with the environment.
Related Classes/Methods:
mani_skill.agents.base_agentmani_skill.agents.controllers.base_controllermani_skill.agents.robots.panda.pandamani_skill.agents.robots.allegro_hand.allegro
Data & Trajectory Management [Expand]
Handles the collection, storage, loading, and replay of simulation trajectories and demonstration data.
Related Classes/Methods:
mani_skill.trajectory.datasetmani_skill.trajectory.replay_trajectorymani_skill.utils.wrappers.record
Contains implementations of various reinforcement learning and imitation learning algorithms for policy training and evaluation.
Related Classes/Methods: