Researchers from Facebook and Georgia Tech have created a digital model of house to train AI.
The contest: A virtual agent is tasked with looking for something in a simulated home after parsing a natural-language question. Questions would be something like “What color is the car?” or “Where is the coffee table?”
Solving the puzzle: Finding the answer requires an agent to understand the question and then explore the virtual space in search of the relevant object. Over time, the program figures out the best areas of the house to explore to get the answer.
Why it matters: The research may feed into more futuristic projects. Future home robots might use AI software that has learned facts like what a bedroom is, or where one might be found by exploring virtual home first.
The contest: A virtual agent is tasked with looking for something in a simulated home after parsing a natural-language question. Questions would be something like “What color is the car?” or “Where is the coffee table?”
Solving the puzzle: Finding the answer requires an agent to understand the question and then explore the virtual space in search of the relevant object. Over time, the program figures out the best areas of the house to explore to get the answer.
Why it matters: The research may feed into more futuristic projects. Future home robots might use AI software that has learned facts like what a bedroom is, or where one might be found by exploring virtual home first.
No comments:
Post a Comment