Carding 4 Carders
Professional
- Messages
- 2,724
- Reaction score
- 1,588
- Points
- 113
NVIDIA robots are now in the game.
NVIDIA Research has introduced Eureka, an artificial intelligence based on GPT-4 that trains robots to perform various tasks using reward algorithms. One of Eureka's most notable achievements has been teaching the robotic arm to do human-level pen-spinning tricks, and as shown in the YouTube video, even better than many of us.
Eureka also trained four-legged robots, dexterous-handed robots, helper robots, and other robots to perform almost 30 different tasks, including opening crates, using scissors, and catching balls. According to NVIDIA Research, Eureka's trial-and-error reward programs are 80% more efficient than programs written by human experts. This resulted in a more than 50% improvement in robot performance. Eureka also self-evaluates based on learning outcomes, adjusting reward functions as needed.
NVIDIA Research has published a library of Eureka algorithms, inviting everyone to test them on the NVIDIA Isaac Gym platform, designed for research in the field of reinforcement learning.
The idea of training robots with other robots is becoming increasingly popular. In May 2023, an article was published in the journal Transactions on Machine Learning Research, which introduces the new SKILL (Shared Knowledge Lifelong Learning) system. The system allows the AI to learn 102 different skills, including diagnosing diseases from chest X-rays and identifying types of flowers.
Artificial intelligences shared their knowledge, acting as teachers, and were able to master each of the 102 skills. Researchers from institutions such as MIT and the University of Bristol have also been successful in using AI to train robots to manipulate objects.
NVIDIA Research has introduced Eureka, an artificial intelligence based on GPT-4 that trains robots to perform various tasks using reward algorithms. One of Eureka's most notable achievements has been teaching the robotic arm to do human-level pen-spinning tricks, and as shown in the YouTube video, even better than many of us.
Eureka also trained four-legged robots, dexterous-handed robots, helper robots, and other robots to perform almost 30 different tasks, including opening crates, using scissors, and catching balls. According to NVIDIA Research, Eureka's trial-and-error reward programs are 80% more efficient than programs written by human experts. This resulted in a more than 50% improvement in robot performance. Eureka also self-evaluates based on learning outcomes, adjusting reward functions as needed.
NVIDIA Research has published a library of Eureka algorithms, inviting everyone to test them on the NVIDIA Isaac Gym platform, designed for research in the field of reinforcement learning.
The idea of training robots with other robots is becoming increasingly popular. In May 2023, an article was published in the journal Transactions on Machine Learning Research, which introduces the new SKILL (Shared Knowledge Lifelong Learning) system. The system allows the AI to learn 102 different skills, including diagnosing diseases from chest X-rays and identifying types of flowers.
Artificial intelligences shared their knowledge, acting as teachers, and were able to master each of the 102 skills. Researchers from institutions such as MIT and the University of Bristol have also been successful in using AI to train robots to manipulate objects.
