For embodied ai specifically in humanoid robots, refer to Embodied AI in Humanoid Robots.
Writed in the Communications of the ACM,
Embodied Artificial Intelligence (EAI) integrates artificial intelligence into physical entities like robots, endowing them with the ability to perceive, learn from, and dynamically interact with their environment.
Foundation Models
Real-World Robot Applications of Foundation Models: A Review lists fundation models can be used in robotic tasks divided by modalities.
Robotic Foundation Models
Refer to Robotic Foundation Models.
Dataset Collection
For the ease of collecting export demonstration datasets with real robots, robot teleoperation systems suitable for collecting dataset have been proposed.
Foundation Models Applications
Low-Level Perception
Representative studies utilizing foundation models, such as CLIPort and REFLECT, to extract semantic information from images, texts and spatial information.
High-Level Perception
Foundation models for high-level perception involve the transformation and utilization of results obtained from low-level perception into forms such as maps, rewards, and motion constrains.
High-Level Planning
Foundation models for high-level planning execute higher-level abstract task planning, excluding direct control.
Low-Level Planning
Foundation models for low-level planning execute low-level motion control, including joint and end effector control.
Data Augmentation
Foundation models augment dataset by changing background, adding distractors and altering object textures to improve the robustness of the learning.
Robotic Tasks
-
Navigation
-
Manipulation
-
Mobile Manipulation (Navigation w/ Manipulation)
-
Locomotion