Recently, a dog has become completely popular.
It is busy on CCTV programs,
busy on business trips across the country,
attending various science and technology exhibitions,
and interacting with audiences from all over the country
. What has this "dog" done, and it has caused such a big response? This will start from 2 months ago.
On September 25 this year, Huawei's Full Connect 2020 Day3, it was Shi Shiran's stage, and the image of the clever and obedient robot dog was vividly displayed.
But if you really think it can only be a boon to others, then it's too small to underestimate its ability. As the ice screen in the center of the stage slowly opened, its real task test room, a chemical plant scene model, was revealed under the dazzling lights, and the test question it faced was to complete chemical plant inspections and deal with crisis tasks.
Task 1: Perceive the crisis
As we activated the fake fire device in the chemical plant model, we simulated an accidental fire scene caused by a gas pipeline leak. The robot dog empowered by HUAWEI CLOUD AI stepped up to the flames at this moment. Facing a situation different from previous inspections on the scene, it stopped, called out the target detection algorithm based on EI-backbone technology, detected and used the yellow frame. Mark out the flame.
Based on this new pre-training model + small sample fine-tuning efficient training mode, it will comprehensively enhance the industry's AI landing ability and experience. Take medical image segmentation as an example. In the past, hundreds or thousands of labeled data were needed for training. With the support of EI-Backbone, only dozens or even dozens of labeled data can be completed, saving more than 90% of labeling costs. .
Task 2: Crisis awareness
The flame was discovered, but how to make the robot dog understand the harm caused by the flame in the chemical plant scene? Through the knowledge calculation solution for hazardous chemical disposal deployed on HUAWEI CLOUD and access to the intelligent question answering application of knowledge calculation, the cause of the fire and the solution of "turn off the gas pipeline switch and cut off the combustibles can extinguish the flame".
Knowledge computing is a new way to realize the combination of industry knowledge and AI . It can extract, express and calculate knowledge from industry experts' experience, massive information data, and traditional mechanism models with the help of a series of knowledge graphs, natural language processing and AI technologies, and then generate A more accurate model releases the power of knowledge and empowers machines and people again.
At present, Huawei cloud knowledge computing solutions are applied in industries such as PetroChina Oil and Gas Exploration, FAW Group, Huaxin Stainless Steel, and urban transportation, reducing costs and increasing efficiency for enterprises, and assisting intelligent upgrades in the industry.
Task 3: Action decision
Since the path planning provided by the knowledge calculation does not correspond to the actual site, the original travel route will go straight through the flame, causing damage to the robot dog.
To this end, by creating a grid world that simulates a chemical plant, using ModelArts reinforcement learning preset algorithms for training, providing a new path for robot dogs to avoid flames.
In the end, the robot dog automatically circumvented the flames and obstacles, pressed the main control button of the gas pipeline, and completed the task of handling the fire crisis.
In the entire task, the realization of AI empowering robot dogs and the completion of the " perception-cognition-decision-making " capability is the ModelArts3.0, which was launched at the Huawei Full Connect Conference .
ModelArts3.0 is an AI development platform for Huawei Cloud AI to be built for the industry. In the past, model selection and hyperparameter adjustment required a lot of expert experience and trial and error cost. Through the full-space network architecture search and automatic hyperparameter optimization technology provided by EI-Backbone , It can be completed quickly without manual intervention, and the accuracy is greatly improved. And through the HUAWEI CLOUD EI backbone model, a high-precision model can be trained with only a small sample data , and the model training of several weeks/months can be compressed to the minute level.
HUAWEI CLOUD ModelArts+ knowledge computing injects wisdom into the robot dog and creates greater imagination for the application of AI in the industry. In the future, what kind of sparks can Huawei Cloud AI and robots burst out? We will wait and see!
Finally, let's take a look at our big stars!
I was turned over by a robot dog?