Sonar Earables and Virtual Rooms Push AI Boundaries
Researchers develop a sonar-based earable that reads facial expressions through sound echoes, while AI2 introduces a framework to generate custom training environments for robots.
The latest AI research highlights a unique "earable" device from Cornell University that uses sonar to read facial expressions. The device resembles bulky headphones and bounces acoustic signals off a wearer's face, with a microphone capturing the tiny echoes produced by moving features like the lips and eyebrows. An AI algorithm then translates these echo profiles into complete facial expressions, offering a potentially sleeker alternative to the massive camera rigs traditionally used in animation.
This innovative earable currently faces a few hardware limitations, including a short three-hour battery life and the need to offload processing to a connected smartphone. Additionally, the system requires 32 minutes of facial data to train the echo-translating algorithm before it can recognize expressions. Despite these hurdles, the technology shows promise for simplifying motion capture in movies, TV, and video games, and it could eventually help animate humanoid robots.
To help those future robots navigate the real world, the Allen Institute for AI introduces ProcTHOR, a framework that procedurally generates custom indoor environments. The system creates thousands of unique scenes, such as classrooms and offices, complete with realistic lighting, varied surface materials, and household objects. Simulated robots use these diverse digital rooms to practice tasks like picking up objects and moving around furniture, exposing them to a wide variety of spaces to improve their real-world navigation skills.