The new company of D-Wave founder wants to combine artificial intelligence, robotics and exoskeletons

Release date: 2016-09-28

Several quantum computing pioneers have recently created a new company, Kindred, to develop advanced artificial intelligence systems for robotic control and training.

If quantum computing isn't enough to burn brains, a founder of D-Wave Systems looks at another future idea: using artificial intelligence and high-tech exoskeletons to make humans (or let monkeys – according to the technology A description file) can control and train the intelligent robot army.

Geordie Rose, co-founder and CTO of D-Wave, is selling a computer that is said to use quantum mechanical effects - it is said that it has reached hundreds of millions of times the speed of traditional computers on specific problems. .

And IEEE Spectrum found that Rose is also the CEO of another company, Kindred Systems (also known as Kindred AI), a stealth-based startup founded in 2014 to provide remotely controlled and autonomous robots. Their goal is to make robot programming faster and cheaper, and may even revolutionize the work of the world.

Image from US patent application US20160243701A1

Kindred has already raised $10 million - according to Data Collective, the venture capital firm has won one of the rounds. Another Silicon Valley venture capital firm, Eleven Two Capital, also participated in the investment. Data Collective describes in a blog post that Kindred is "using artificial intelligence-driven robots that enable a human worker to do four people's work."

Kindred recently filed a US patent describing a system in which an operator can wear a head-mounted display and an exoskeleton set to perform daily tasks. Data from the kit and other external sensors can then be used to control the remote robot after being analyzed.

Kindred has always been low-key, has not published media press releases, and has only a very simple website: However, in November last year, a former D-Wave researcher, Kindred co-founder and CTO told some technology professionals that the company was building a personal robot that uses machine learning to identify patterns and make decisions.

She said: "Quantum mechanics is cool, but the humanoid intelligence in robots is even cooler."

Kindred, based in Vancouver, British Columbia, Canada, recently published a US patent application that reveals its ambitions. This document describes a head mounted display and an exoskeleton system with sensors and actuators that the operator can wear to perform everyday tasks. The cloud computer then analyzes the data from the kit and other external sensors and controls the remote robot based on the results of the analysis. These data can also be used to train machine learning algorithms, which allows the robot to automatically mimic the operator's actions.

"Operators may include non-human animals, such as monkeys," the patent writes. "And the operator interface may... need to be resized to bridge the gap between human operators and monkey operators." (In fact, This is not the first device that allows monkeys to directly control the robot, but the previous results are more concerned with the brain-computer interface (see reading Stanford to develop machine learning brain-computer interfaces to help monkeys play Shakespeare's famous sentences), rather than robot control and automation. )

From US patent application US20160243701A1. Exoskeleton rendering with sensors and actuators, the operator can wear it to remotely control the robot

This patent application describes some details about the operator interface, a set of wearable robot kits containing head and neck motion sensors (top), equipment for capturing arm movements and tactile gloves. The operator can use the foot pedal to control the movement of the robot and use a virtual reality headpiece like the Oculus Rift to experience what the robot sees. The kit even includes chemical and biosensors, as well as EEG and MRI equipment for capturing brainwave signals.

The robot it envisions is a 1.2-meter-high humanoid robot with an outer layer of synthetic skin, two (or more) robotic arms (used as a hand or gripper), and a wheeled chassis for movement. The camera mounted on its head transmits high-precision video to its monkey operator, along with sensors for infrared and UV imaging, GPS, touch, proximity and strain detection, and even a radiation detector .

The system can be used for direct remote control, and operators can operate remote robots to perform industrial and home tasks. The legendary robotics lab, Willow Garage, experimented with a similar system, Heaphy, in 2011, with some success.

Now Kindred wants to implement the next step for a remote robot. "Although a large amount of information contained in the human brain to perform various human-executable tasks is available, robot-related devices used to perform these tasks have not been or are not fully utilized." The patent document stated. In addition, it mentions: "Operators may also contain non-human animals, such as monkeys. And the operator interface may... need to be resized to bridge the gap between human operators and monkey operators."

More importantly, the company wanted the system to learn from its operators and ultimately allow the robot to perform tasks without the control of humans – or monkeys. The patent states: "Device control commands and environmental sensor information generated during multiple rounds of operation can be used to derive automatic control information that can be used to facilitate automated behavior in automated equipment."

The document also indicates that Kindred will use "deep hierarchical learning algorithms" to achieve this goal, including the conditional deep belief network (CDBN) and the conditionally constrained Boltzmann machine (CDBN: conditional deep belief network) CRBM: conditional restricted Boltzmann machine) - This is a powerful recurrent neural network.

From US Patent Application US20160243701A1, the above diagram shows how multiple operators and robots communicate via Kindred's cloud-based artificial intelligence system.

In fact, Graham Taylor, one of the inventors of the patent, has some research contributions on CDBN and CRBM. Taylor is the head of the Machine Learning Research Group at the University of Guelph in Ontario. He studied undergraduate Geoff Hinton at the University of Toronto (Hinton co-invented the Boltzmann machine in 1985, and he now works at both Google and the University of Toronto).

The quantum computing company D-Wave said that the system's operation is "similar to the Boltzmann machine", and its research team "is studying the parallel use of these architectures to fundamentally accelerate the learning of deep and hierarchical neural networks."

In 2010, Geordie Rose co-authored a paper entitled "The Ising model: teaching an old problem new tricks" that quantum computers can be more efficient than traditional computers in some types of machine learning applications. Will this be the beginning of a new field, quantum robotics?

Neither Kindred nor D-Wave responded to IEEE Spectrum's request for comment, but according to LinkedIn and Canadian government records, Kindred has approximately 25 employees in Vancouver, including several former D-Wave employees. The company appears to be in the San Francisco Bay Area as well, including a mechatronics engineer specializing in mechanical design and mechatronic integration.

As for practical applications, the patent refers to industrial manufacturing, household chores and even entertainment. It said: "The tasks it performs may include making a cup of coffee or performing a dance. An operator...maybe a performer... can provide a recordable set of actions (such as a set of verbal communication via a robot) The speaker on it plays out.)

Although it is unclear how much Kindred's remote robotic system has been done, the document has given the system's exoskeleton 3D rendering model, details of some components, and photos of the glove assembly and robotic track.

The robotics experts at IEEE Spectrum say this makes programming robots simpler, and as Kindred wants to do, it will be a big step forward in the field. But there is some doubt as to whether the company has the ability to deliver the system described. Tim Filed, who developed the Heaphy Remote Robot System at Willow Garage, said, "The portion of applied machine learning is far beyond today's top level, and the amount of data it needs is astronomical." He mentioned that the Google Research system allows robots to learn from Eight hundred thousand attempts were made to select objects in the basket. He explained, "Imagine the time it takes to use manual operations for 800,000 attempts, which is now impossible."

Oregon State University robot expert Bill Smart said, "This is a good idea in theory, but the trick is to grasp the task environment. In addition, I bet that humans can't make the robot achieve the best way of exercising. Because people and machines have different dynamics."

How do you use primates to teach robots how to sing and dance? Smart joked, "If you have an unlimited number of monkeys, would you get an optimal controller? But keeping them task is a nightmare. "

Source: Robot Network

Shrimp Products

Shelled shrimps,Dried shrimps,Frozen Bamboo shrimp,Pandalus borealis,Red Shrimp,Coldwater Shrimp

Zhejiang ocean family co.,ltd , https://www.ocean-family.com