Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a groundbreaking system called Neural Jacobian Fields (NJF) that revolutionizes how robots are controlled.
Instead of relying on complex mathematical models tailored to rigid and expensive robotic structures, NJF enables robots to autonomously learn about their own bodies and responses to commands using only visual input.
During the training process, a robot performs random movements while multiple cameras observe its actions. The NJF system builds an internal model of the robot’s physics, mapping control signals to actual movements. This approach builds on advancements in Neural Radiance Fields (NeRF) technology, adapting it to capture the dynamic behavior of a robot’s physical form.
The key innovation lies in its simplicity: after training, the robot can be controlled in real time using just a single standard camera. This eliminates the need for costly sensors or intricate pre-programmed models. The system has been successfully tested on a range of devices, from soft pneumatic manipulators to standard 3D-printed structures, demonstrating its versatility.
Also read:
- Microsoft’s Copilot to Gain Personalization with Expressive Avatars and Conversational Memory
- Google Launches Opal: A Visual Builder for Creating and Publishing AI-Powered Apps
- 4chan Hacker Leaks 59GB of Tea App User Data, Including Photos, Locations, and Driver’s Licenses
While NJF currently lacks tactile feedback, it paves the way for creating more affordable and adaptable robots. This technology enables robots to adjust to their own unique, even unconventional, physical designs, opening new possibilities for flexible and cost-effective robotics in the future.
