Use Artificial Intelligence to replicate the functioning of the brain
A robotic game of cat and mouse shows how brain-inspired AI chips with low power consumption can be applied to autonomous driving, the Internet of Things and smart home devices.
Researchers at Tsinghua University in China have robotized the cat-and-mouse game, making a robot cat chase and harass a robot mouse.
The llama cat robot tianjicat and has a computer chip inspired by the brain (neuromorphic), called TianjicXwhich executes various Artificial Intelligence instructions to chase the robot mouse, which is configured to move randomly within a space with obstacles.
The task, from a robotics point of view, is not an easy one, and to make it work, Tianjicat had to track the robot mouse using visual recognition and sound detection.
With this information, he had to “conceive & rdquor; how to chase the robot mouse during its movements and calculate the best way to intercept it, avoiding different obstacles.
In the end, it succeeded, and the researchers note that the TianjicX chip halved the amount of power required for the robot cat to make decisions during the chase, compared to a chip for AI computing designed by NVIDIA.
neuromorphic example
The researchers note that this is an example of experiments with neuromorphic systems that could allow small robots to make decisions using limited computing resources and power.
In real life, autonomous mobile robots often have limited amounts of power to move and make decisions. This limitation becomes a challenge during search and rescue missions as well as exploratory research. The new development has overcome this challenge.
“We were eager to design a neuromorphic chip that would function as the brain of robots, enabling it to have powerful intelligent processing capabilities and to cope with complex and unknown environments,” he says. Luping Shione of the study’s authors, cited by the AAAS.
Real-time multitasking
AAAS stresses that robots are slowly getting into the realm of real-time multitasking, but before they can truly cross that bridge, their hardware systems need to be upgraded.
“The multitasking intelligent robot needs high computing power, high concurrency, low power consumption, high resource scheduling flexibility, and easy-to-use computing hardware to run intensive algorithms locally in real time,” the authors of the new development point out. .
At this time, processing units cannot efficiently meet AI needs for inexpensive, simultaneous, and adaptive processing. But neuromorphic systems, which approach computation in a more cooperative way, could be a solution, they add.
new neuromorphic system
to invent a new neuromorphic computing system that can support neural networks holistically, they created a model called Rivuletbased on how the different parts of the human brain work together to complete many tasks at once.
They were inspired by the features of concurrency and cooperation in multiple brain regions, to propose an execution model that allows multiple neural network tasks to schedule resources on the chip efficiently, flexibly, and simultaneously.
Using Rivulet, the authors built the TianjicX electronic chip, as well as additional software. They then incorporated both into an autonomous mobile robot called Tianjicat and ran it through a cat-and-mouse demo game.
cat and mouse
The game of cat and mouse requires the collaboration of multiple sensory information to deal with complex and dynamic scenarios, which is considered key for robots to achieve human-like intelligence, the protagonists of this development point out.
Throughout the demo, the neuromorphic chip and its supporting software helped the robot perform intense multitasking. Further analysis revealed that TianjicX halved the amount of energy Tianjicat consumed while processing information, and significantly reduced the delays between making decisions and subsequent action.
According to the researchers, these findings show how their platform could be used in AI research and real-world scenarios.
We provide an exploration platform for AI researchers and promote the study of brain-inspired Artificial Intelligence.
computing scenarios
“Our platform also has great potential to be applied in edge computing scenarios due to its multitasking processing capabilities, such as autonomous driving, Internet of Things, and smart home devices,” they conclude.
Neuromorphic systems have yet to be commercialized on a large scale, but their relatively low size, weight and power requirements could provide practical advantages for robotic deployment.
Reference
Neuromorphic computing chip with spatiotemporal elasticity for multi-intelligent-tasking robots. Xjing Pei et al. Science Robotics, 15 Jun 2022; Vol 7, Issue 67. DOI: 10.1126/scirobotics.abk2948