Embodied artificial intelligence is the branch of AI that deals with agents that can sense, navigate, and interact with the real world. This is a key component of robotics applications.
One of the key limits of embodied AI is training reinforcement learning agents. Reinforcement learning agents used in robotics must obtain their experiences in the real world. This poses severe timing and safety constraints on the training of RL agents.
In the past few years, several efforts have been led to create simulated environments for training RL agents on embodied AI tasks. But replicating the dynamics of the real world is a very very difficult—if not impossible—task.
A recent challenge developed by scientists at IBM, the Massachusetts Institute of Technology, and Stanford University, aims to push the limits of embodied AI by providing an environment that is visually and physically realistic. Titled “ThreeDWorld Transport Challenge,” the test is meant to push the limits of embodied AI and usher new innovations in the field.
For more on robotics and AI: