Researchers build hybrid chip able to run autonomous bicycle

Researchers build hybrid chip able to run autonomous bicycle

A team made up of members from a host of institutions in China, one in Singapore and one in the U.S., has built a hybrid chip that can control an autonomous bicycle. In their paper published in the journal Nature, the group describes the effort that led to the chip and how well it worked when tested.

In modern computer science, there are two basic types of ongoing research—one involves the traditional binary approach—the other involves trying to get machines to behave like the . In most cases, the two approaches do not really go together because of communication difficulties between the two systems. But that may change as the team working in China has found a way to create not just a way for two such systems to communicate seamlessly, but to do it on an actual chip—one that works as demonstrated by its ability to control an autonomous bicycle. They call the new chip Tianjic and it has what they describe as an FCore architecture. Tianjic has 156 FCores, all speaking to one another in binary. Together the FCores were able to carry out processing using 40,000 compute units.

Because the chip allows easy between its networks, it is able to use the advantages of both types of them—a necessity for keeping a bicycle balanced while moving along a course. But the chip was able to do more than that—it also was able to carry out obstacle avoidance and could respond to oral commands.

Credit: Pei et al., Nature

Intriguingly, the networks running on the chip were vastly different—one was based on calculating things like distance and speed. Another was based on spiking communications, a model based on the way neurons in the brain communicate information and use it to process and respond to real-world conditions. And was carried out by yet another network—a convolutional neural network that was similar to those in use in some commercial applications.

In summing up their accomplishments, the researchers added another interesting note: they suggest that the creation of their new AI is likely to stimulate Artificial General Intelligence (AGI) development. AGI is a term used to describe artificial intelligence that is on a par with human .

Explore further

An all-optical neural network on a single chip

More information: Jing Pei et al. Towards artificial general intelligence with hybrid Tianjic chip architecture, Nature (2019). DOI: 10.1038/s41586-019-1424-8
Journal information: Nature

© 2019 Science X Network

Citation: Researchers build hybrid chip able to run autonomous bicycle (2019, August 1) retrieved 14 October 2019 from
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.

Feedback to editors

User comments

Aug 01, 2019
Next - luggage that follows you around the airport and fetches itself from the carousel-

Aug 01, 2019
Next - luggage that follows you around the airport and fetches itself from the carousel-
How about one that follows me home with my groceries.

Aug 02, 2019
Reminds me about April fools prank from Amsterdam 3 years ago > https://www.youtu...PNwZex9s > Welcome to the future :)

Aug 02, 2019
a necessity for keeping a bicycle balanced while moving along a course

The self-balancing bicycle is a trivial problem that can be solved with well-tuned PID controllers and good gyroscope sensors, because it's a relatively simple physical system (a modified inverted pendulum). Getting the device to turn and change course simply requires altering the set-point values of the PID controller so the bike leans to a certain angle that results in a certain rate of turning. (you can program in the counter-steer sequence to get the initial lean) It doesn't actually need fancy multi-core processors. Neither does the simple object recognition and avoidance.


The whole thing could be run on a Raspberry Pi

Aug 02, 2019
The angle of the front wheel along with the forward speed of the bike determines the force that is balancing the inverted pendulum. Knowing that, the Wikipedia article describes the rest of the control algorithm from there on in a simple diagram:


In the case of this simple control system being applied to the bicycle, altering the "command position" value by a fixed amount results in the bicycle turning a fixed amount and then returning upright. This abstracts the control of the bicycle in a way that a simple image recognition system can point the bicycle to where there are no obstacles simply by changing the variable that represents the command position.

Notice also that this balancing algorithm can be implemented entirely in the analog domain with a handful of op-amps as well. It doesn't actually need a computer, much less a neural network. This would be cutting edge in 1970's.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more