Artificial intelligence in robotics
Previously in this series, we saw that NVIDIA (NVDA) is working on several components to make AI (artificial intelligence) cars reality. Another area in which the company is leveraging AI is robotics. At the 2017 Computex conference, NVIDIA CEO Jensen Huang announced the Isaac Initiative, a robotic platform that accelerates development of other robots and mechanical devices.
At present, robots are programmed by humans to perform a particular task. With Isaac, a robot trains another robot, with no human intervention. The Isaac platform is a combination of NVIDIA’s Jetson TX2 hardware, Astro AV Software Stack, Isaac Training Lab, and several open-source platforms.
How does Isaac work?
Isaac uses AI algorithms to train reinforcement learning, which creates a virtual brain. It then downloads this virtual brain into a Jetson TX2 GPU (graphics processing unit) to create a new robot. This robot, pre-trained at the Isaac Training Lab, then trains itself on the domain in the physical world.
The most important component in Isaac is developing a virtual brain. For this step, NVIDIA uses the Isaac Training Lab and Astro AV Stack. Astro, NVIDIA’s API (application programming interface), enables a device to perceive and move around environments.
The Isaac Training Lab
The Isaac Training Lab creates photorealistic, simulated environments that follow the laws of physics, which help a program translate its learning from the real world into a physical robot. At the Computex conference, Jensen Huang demonstrated how a program taught itself to play hockey in the virtualized Isaac Training Lab. NVIDIA demonstrated a similar training platform, Project Holodeck, which creates a virtual world in which to train AI and humans.
Automaker Toyota (TM) and drone maker Teal will be the first clients of NVIDIA’s Isaac Initiative. Toyota will use Isaac to build service robots. While these virtual training platforms may not have any immediate impact on consumers, they could build a strong foundation for NVIDIA to develop future robots and stay ahead of Intel (INTC) in the field. Next, we’ll look at NVIDIA’s initiative to build AI cities.